Pith. sign in

REVIEW 4 major objections 6 minor 3 cited by

Embodied VideoAgent: Persistent Memory from Egocentric Videos and Embodied Sensors Enables Dynamic Scene Understanding

T0 review · 4 major / 6 minor · reviewed 2026-08-10 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read A memory agent that fuses egocentric video with depth and pose into a persistently updated object memory beats video-only models on dynamic 3D scene understanding (gains up to 11.7%).

desk verdict Solid extension of VideoAgent with a fair VQ3D result, but the OpenEQA headline gain is undermined by a subset-vs-full comparison and missing ablations. read the letter →

arxiv 2501.00358 v2 pith:CS5HOXXP submitted 2024-12-31 cs.CV

Yue Fan , Xiaojian Ma , Rongpeng Su , Jun Guo , Rujie Wu , Xi Chen , Qing Li This is my paper · ORCID
classification cs.CV
keywords egocentricvideopersistentobjectmemoryembodiedsceneunderstandingmultimodalagentdepthandcameraposeVLMupdate3Dlocalizationquestionanswering
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper proposes Embodied VideoAgent, an LLM-based agent that answers questions about dynamic 3D scenes seen from an egocentric camera. Its central claim is that a persistent object memory built from video plus depth maps and camera poses, and updated automatically by a vision-language model whenever an action changes an object, gives large gains over both end-to-end video-language models and earlier video-memory agents. On three benchmarks the agent reports gains of 4.9% on Ego4D-VQ3D 3D localization, 5.8% on OpenEQA embodied question answering, and 11.7% on EnvQA long-form interaction question answering. A sympathetic reader should care because the result would mean that structured, updatable memory, not just a larger model, is what makes embodied scene understanding work in changing scenes.

What carries the argument

The load-bearing mechanism is the persistent object memory: a table of object entries, each holding an object id and category, a state description, relations to nearby objects computed from 3D bounding boxes, a 3D bounding box obtained by 2D-3D lifting using depth and camera pose, and visual features of both the object crop and its surrounding frame. Construction pipelines an open-vocabulary detector into 2D detections, lifts them to 3D boxes, and merges duplicates through a two-stage re-identification procedure that separates static from dynamic objects, using intersection and volume-similarity scores for the former and a visual-similarity threshold for the latter. The companion mechanism is the vision-language-model-based memory update: when the agent perceives an action, it renders each candidate object's 3D box onto the current frame and prompts a vision-language model to decide which memory entry is the action's target, then changes that entry's state and logs the action in a history buffer. This combined memory-plus-update structure is what the paper credits for precise temporal and spatial localization and for answering event, order, and state questions in dynamic scenes.

What would settle it

Running Embodied VideoAgent and the exact same baselines on the full OpenEQA validation set, or ablating the VLM-based memory update and checking whether EnvQA Event and Order accuracies drop, would settle whether the reported gains come from the persistent memory and update mechanism.

Watch

Extended reading notes

Core claim

Embodied VideoAgent's central discovery is that fusing egocentric RGB with embodied sensor readings (depth and camera pose) into a persistent per-object memory, and refreshing that memory when actions occur, enables a tool-calling agent to track objects in dynamic scenes more accurately than either end-to-end multimodal LLMs or the video-only memory baseline. Each memory entry holds an object id and category, a state label ("normal", "open", "close", "in hand"), related objects derived from 3D bounding boxes, a 3D bounding box lifted from 2D detections via depth and pose, and visual features of the object crop and its surrounding context. Re-identification merges new detections into existing entries using visual and spatial similarity scores, with separate handling for static and dynamic objects. When an action annotation indicates an activity over an object, the agent visually prompts the vision-language model with rendered 3D boxes to identify the target entry and programmatically updates its state, logging the action in a history buffer. In evaluation, this design yields the reported gains: 85.37% success on Ego4D-VQ3D, 47.0% on the OpenEQA subset, and 25.91, 68.00, and 35.50 on EnvQA event, order, and state categories.

Load-bearing premise

The headline OpenEQA gain assumes that the randomly selected one-fifth subset of the benchmark behaves like the full set, so that baseline scores borrowed from the full benchmark remain valid for comparison.

Editorial extensions

If this is right

  • Dynamic 3D scene understanding can be improved by explicit persistent memory rather than by scaling end-to-end video models, because the agent outperforms both on all three benchmarks.
  • Egocentric video alone is not sufficient: depth and camera poses contribute the geometry needed for reliable object tracking and 3D localization.
  • Associating perceived actions with target objects through vision-language-model visual prompting is a key ingredient for event and order understanding, where the paper reports the largest relative gains.
  • The same memory supports embodied behavior beyond question answering: persistent entries let a robot retrieve an occluded object, such as an apple hidden behind a box, and complete a pick-and-place task.
  • The agent architecture can generate synthetic user-assistant interaction episodes by having one LLM propose tasks and the memory-augmented agent execute them in a simulator.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Editorial extension: the claimed 5.8% OpenEQA gain would be on firmer ground if the full benchmark were scored with identical baselines; the paper's random one-fifth subset makes the comparison with full-set borrowed baselines an assumption about representativeness.
  • Editorial extension: if the memory update is the true cause of the EnvQA gains, then disabling the VLM-based update while keeping all other components should substantially reduce Event and Order accuracy; this ablation is not reported in the paper.
  • Editorial extension: the method's robustness to noisy poses suggests a natural testable extension of replacing ground-truth sensor inputs with monocular depth and pose estimators end-to-end, then measuring the same three benchmarks.
  • Editorial extension: the persistent memory design could be carried into long-horizon robot deployments where the same object is revisited over hours, a use case the paper illustrates only in short manipulation scenes.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 6 minor

Summary. The paper proposes Embodied VideoAgent, an LLM-based multimodal agent that extends the prior VideoAgent system by constructing a persistent object memory from egocentric video, depth maps, and camera poses, and by updating that memory with a VLM when actions or activities are perceived. The agent also maintains action and visible-object history buffers and uses four perception tools plus seven embodied action primitives. It is evaluated on Ego4D-VQ3D object localization, OpenEQA embodied question answering, and EnvQA video question answering, with reported gains of 4.9%, 5.8%, and 11.7% respectively, and is further demonstrated on synthetic embodied interaction generation and a robot manipulation scenario. The central claim is that the persistent memory architecture, rather than the underlying LLM/VLM backbone, is responsible for the improvements.

Significance. If the empirical claims hold, the paper would demonstrate a practical and cost-effective way to handle dynamic 3D scene understanding by augmenting an LLM agent with structured persistent memory and VLM-based memory updates. The manuscript has several strengths: the VQ3D evaluation shares precomputed camera poses and depth maps across methods, making that comparison clean; the appendix provides detailed algorithms (object re-ID, memory update, tool prompts) and honest failure cases; and the application demonstrations, while qualitative, are suggestive of downstream utility. However, the headline OpenEQA number is not established because it compares a one-fifth subset score against full-benchmark baselines, and no ablation isolates the memory contribution from the stronger GPT-4o/InternVL2-8B backbone. The EnvQA and VQ3D results also lack statistical grounding. These issues are load-bearing for the paper's central claims, though they appear fixable with additional experiments.

major comments (4)
  1. [Section 3.2.1, Table 2] The abstract's claim of a 5.8% gain on OpenEQA is not supported by the reported experiments. Embodied VideoAgent is evaluated on a randomly selected one-fifth subset of OpenEQA, while the GPT-4 w/LLaVA-1.5 (43.6 ALL) and GPT-4 w/CG (36.5 ALL) rows are borrowed from full-benchmark evaluations. The paper argues the subset is harder based on Video-LLaVA and LLaMA-VID drops, but this does not make a subset score comparable to a full-set score for different baselines. If the intended comparison is 47.0 (subset) against 43.6 (full set), the claimed 5.8% gain is not established; if the 5.8% refers to the gap between the GPT-4o and InternVL2-8B variants (47.0 vs 41.2), then it is not a gain over an external baseline. Please evaluate all baselines on the exact same subset, or run Embodied VideoAgent on the full OpenEQA set, and report per-split counts and standard errors.
  2. [Section 3.2.2, Table 2] No ablation isolates the contribution of the persistent object memory, VLM-based memory update, and history buffers from the choice of backbone models. Embodied VideoAgent (GPT-4o) differs from VideoAgent not only in the memory design but also in using GPT-4o as the LLM and InternVL2-8B for VQA, so the observed gains over VideoAgent could be due in part to the stronger backbone. Please provide an ablation that either runs VideoAgent with the same GPT-4o/InternVL2-8B tools or removes the persistent memory and update mechanism from Embodied VideoAgent while keeping all other components fixed. This is necessary to support the paper's attribution that 'the strong performance of Embodied VideoAgent mainly attributed to the precise frame localization using consistent object memory and history buffer.'
  3. [Section 3.2.3, Table 3] The EnvQA results are reported as per-category accuracies on 200 questions per type with no variance or significance testing, and the abstract's 11.7% gain is not derivable from the table. The largest gain is in Events (25.91 vs 10.19 for Video-LLaVA), but the Orders gain over VideoAgent is only 2.5 points (68.0 vs 65.5), which is within likely noise for 200 questions. Please state the aggregation rule used to compute the 11.7% (e.g., macro-average over the three categories) and provide confidence intervals, a paired test, or a bootstrap estimate. Without this, the 'significant performance gain' is not quantitatively supported.
  4. [Section 3.1, Table 1] The VQ3D headline is a 4.88-point absolute gain in Succ% (85.37 vs 80.49) on a single validation set. Given that the validation set is small (QwP% values imply roughly 80-90 queries), this difference could correspond to only a few queries. Please report the number of queries and provide a bootstrap confidence interval or a McNemar test to establish that the gain is not due to chance. As it stands, the '5%' improvement is not statistically grounded.
minor comments (6)
  1. [Section 3.1.2] The text says Embodied VideoAgent (image) surpasses EgoLoc by 5%, but Table 1 shows 85.37 vs 80.49, which is 4.88%; please use the exact value or say 'approximately 5%'.
  2. [Section 2.2, Note on camera poses] The note states that camera poses for Ego4D-VQ3D are estimated using COLMAP, but Section 3.1.1 says all VQ3D methods use the same precomputed poses and depth images provided by EgoLoc. Please clarify whether EgoLoc's poses are the COLMAP poses, or whether the COLMAP remark applies only to a different experiment.
  3. [Section 3.2.2] The sentence 'validated by the better performance of Embodied VideoAgent (+16.4% over Video-LLaVA on subset) over GPT-4 w/CG (+4.4% over Video-LLaVA on full set)' appears to reference GPT-4 w/CG, but the +4.4% value corresponds to GPT-4 w/LLaVA-1.5; GPT-4 w/CG actually scores below Video-LLaVA. Please correct this attribution.
  4. [Figure 12 caption] The caption says 'placing the lamb on a table' in the context of a lamp object; this should be 'lamp'.
  5. [Throughout] The paper uses both 'Embodied VideoAgent' and 'E-VideoAgent' interchangeably. Define the abbreviation at first use and use it consistently.
  6. [Abstract and Tables 1-3] The abstract's numerical gains (4.9%, 5.8%, 11.7%) are not all directly traceable to the tables; please provide the exact formulas or comparisons used for each number.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the paper's claims are empirical benchmark results, not derivations; no fitted input is renamed as a prediction and no load-bearing argument reduces to a self-citation.

full rationale

Embodied VideoAgent is a system paper. Its central claims are benchmark scores on Ego4D-VQ3D, OpenEQA, and EnvQA, and the construction is an engineering pipeline rather than a derivation chain. The object memory is built from video, depth, and poses with off-the-shelf detectors (YOLO-world, SAM-2, CLIP/DINOv2) plus hand-set thresholds (0.2, 0.45, 0.7), but none of these components is fitted to or defined in terms of the benchmark answers. The re-ID visual similarity coefficients (0.15/0.85) and the VideoAgent design are inherited from the authors' prior work [7], yet this is a transparent reuse of a component, not a self-citation that supplies the paper's conclusion: the new results are measured against EgoLoc, VideoAgent, Video-LLaVA, LLaMA-VID, and GPT-4 baselines on held-out benchmarks. The VLM-based memory update and action buffer use action annotations and visual prompting to update STATE fields; answering EnvQA questions from that buffer is the system's intended operation, not a tautology. The one methodological weakness--comparing a one-fifth OpenEQA subset against full-benchmark borrowed baselines--undermines the statistical force of the 5.8% OpenEQA gain, but that is an evaluation-validity concern, not circularity: the subset scores are not constructed from the baseline numbers. No equation in the paper reduces a target quantity to its own input, and no self-citation is load-bearing in place of evidence.

Assumptions & free parameters 5 free parameters · 5 assumptions · 2 invented entities

The central claim rests on hand-set re-ID thresholds, several pretrained models used as black boxes, the assumption that noisy depth and pose still yield usable 3D boxes, and the representativeness of a one-fifth OpenEQA subset. None of these are derived from first principles, and several are shown by the paper's own failure cases to be brittle.

free parameters (5)
  • Visual similarity weights = 0.15 and 0.85
    Fixed weighting of CLIP and DINOv2 similarity for object re-ID (Eq. 1, Appendix C.1); the memory matching depends on this balance.
  • Static re-ID threshold = 0.2
    Spatial IoU or MaxIoS threshold in Algorithm 1 for matching static objects; chosen by hand.
  • Dynamic re-ID thresholds = 0.7 and 0.45
    Volume similarity and visual similarity thresholds in Algorithm 2 for matching dynamic objects; chosen by hand.
  • Dynamic-object detection visual threshold = 0.45
    Threshold in Appendix C.3 for deciding an object is no longer where it should be and should be marked dynamic.
  • Moving average window sizes = 10 static, 2 dynamic
    Window sizes for merging re-identified objects into memory; static objects update slowly, dynamic objects update quickly.
assumptions (5)
  • domain assumption Pretrained perception models (YOLO-world, SAM-2, CLIP, DINOv2, InternVL2, GPT-4o, LaViLa) provide sufficiently accurate detections, masks, features, and annotations for the memory pipeline.
    Sections 2.2 and Appendices B-D rely on these models without fine-tuning; failure cases in Appendix H show this assumption can fail.
  • domain assumption 2D-3D lifting from depth and camera poses yields 3D bounding boxes accurate enough for re-ID and relation extraction.
    Sections 2.2 and Appendix B depend on this; Appendix H.2 shows a 2D-texture chair fails lifting and leads to a wrong answer.
  • domain assumption Camera poses and depth maps are available or estimable.
    The system requires depth and 6D pose inputs; robustness to noisy poses is tested only on a subset in Appendix E.
  • ad hoc to paper The one-fifth OpenEQA subset is representative of the full benchmark.
    Section 3.2.1 introduces the subset for cost reasons and Table 2 compares subset scores against full-set baseline scores.
  • domain assumption LaViLa action annotations every two seconds and VLM association correctly identify action target objects.
    Appendix D describes this mechanism; EnvQA success relies on it despite the simulated environment lacking hands.
invented entities (2)
  • Persistent object memory entry
    purpose: Stores per-object state, 3D bounding box, visual features, and relations to support dynamic scene understanding.
    This is the paper's core new representation; its efficacy is demonstrated only through the paper's own benchmark results, with no external validation.
  • Action and visible object history buffers
    purpose: Log past actions and detected objects with timestamps to support temporal queries and memory updates.
    Introduced as auxiliary memory components and evaluated only within this paper's system.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Embodied VideoAgent: Persistent Memory from Egocentric Videos and Embodied Sensors Enables Dynamic Scene Understanding." pith.science (2026). https://pith.science/paper/CS5HOXXP

@misc{pith2026250100358,
  author       = {Pith},
  title        = {Pith review of: Embodied VideoAgent: Persistent Memory from Egocentric Videos and Embodied Sensors Enables Dynamic Scene Understanding},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/CS5HOXXP}},
  note         = {Machine review of arXiv:2501.00358}
}
read the original abstract

This paper investigates the problem of understanding dynamic 3D scenes from egocentric observations, a key challenge in robotics and embodied AI. Unlike prior studies that explored this as long-form video understanding and utilized egocentric video only, we instead propose an LLM-based agent, Embodied VideoAgent, which constructs scene memory from both egocentric video and embodied sensory inputs (e.g. depth and pose sensing). We further introduce a VLM-based approach to automatically update the memory when actions or activities over objects are perceived. Embodied VideoAgent attains significant advantages over counterparts in challenging reasoning and planning tasks in 3D scenes, achieving gains of 4.9% on Ego4D-VQ3D, 5.8% on OpenEQA, and 11.7% on EnvQA. We have also demonstrated its potential in various embodied AI tasks including generating embodied interactions and perception for robot manipulation. The code and demo will be made public.

Figures

Figures reproduced from arXiv: 2501.00358 by the authors.

Figure 1
Figure 1. Embodied VideoAgent is a multimodal agent that 1) builds scene memory from both egocentric video and embodied sensory input; 2) utilizes multiple tools to query this memory; 3) activates embodied action primitives to interact with the environments, ef￾fectively fulfills various user requests. habited characters [5, 11, 33]; 3) Maintaining a persistent memory about the scene that allows frequent update over time [12,… view at source ↗
Figure 2
Figure 2. An overview of Embodied VideoAgent. Left: We first translate the egocentric video and embodied sensory input (depth maps and camera poses) into structured representations: persistent object memory and history buffer. While the memory can be updated using VLM to support dynamic scenes where actions are being performed constantly; Right: the LLM within Embodied VideoAgent is prompted to fulfill the user’s request by i… view at source ↗
Figure 3
Figure 3. Visualization of the entries in persistent object memory MO. Each 3D bounding box corresponds to an entry in the memory. As the video proceeds, objects (e.g. the large canned tomato paste) can be tracked/re-IDed and have their memory entries updated [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (12 more)
Figure 4
Figure 4. Figure 4: An illustration of our VLM-based memory update method. [PITH_FULL_IMAGE:figures/full_fig_p004_4.png]
Figure 5
Figure 5. Figure 5: An overview of our synthetic embodied data collection [PITH_FULL_IMAGE:figures/full_fig_p005_5.png]
Figure 6
Figure 6. Figure 6: An episode of generated embodied user-assistant interaction. The episode is produced by the framework mentioned in [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]
Figure 7
Figure 7. Figure 7: Our persistent object memory enables effective real-world robotic manipulation. Using [PITH_FULL_IMAGE:figures/full_fig_p007_7.png]
Figure 8
Figure 8. Figure 8: Frame No.190 that contains blinds are successfully located by [PITH_FULL_IMAGE:figures/full_fig_p017_8.png]
Figure 9
Figure 9. Figure 9: In Section F.2, The frame depicting the coffee machine named "Mr. Coffee" is successfully retrieved by frame_localization. 18 [PITH_FULL_IMAGE:figures/full_fig_p018_9.png]
Figure 10
Figure 10. Figure 10: The objects that are most relevant to "cup" are successfully retrieved by retrieve_objects_by_appearance in [PITH_FULL_IMAGE:figures/full_fig_p020_10.png]
Figure 11
Figure 11. Figure 11: An example of interaction data, which is detailed in [PITH_FULL_IMAGE:figures/full_fig_p023_11.png]
Figure 12
Figure 12. Figure 12: When Embodied VideoAgent fails to accomplish the initial task assigned by the user (in this case, Embodied VideoAgent cannot find the candy bar after searching the entire apartment), the user agent will interactively adjust the task. Finally, Embodied VideoAgent compl…
Figure 13
Figure 13. Figure 13: is an example that demonstrates the thought of Embodied VideoAgent during the episode. The user agent first initializes the task as setting up a cozy corner by placing a cushion on a table. Embodied VideoAgent finds the cushion after a simple search and plans to place…
Figure 14
Figure 14. Figure 14: Due to poor image quality caused by scene reconstruction in HM3D, false object detections happen in frame 13 and frame 30. [PITH_FULL_IMAGE:figures/full_fig_p028_14.png]
Figure 15
Figure 15. Figure 15: The 2D bounding box of the detected chair is not successfully 3D-lifted since it is a 2D texture in the scene. [PITH_FULL_IMAGE:figures/full_fig_p029_15.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 3 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. StateScribe: Towards Accessible Change Awareness Across Real-World Revisits

    cs.HC 2026-04 unverdicted novelty 6.0 of 10

    StateScribe uses a dual-layer memory architecture for episodic scenes and object-centric changes to deliver live and historical descriptions, achieving 83.1% F1 accuracy across revisits in evaluations and user studies...

  2. Move to Understand a 3D Scene: Bridging Visual Grounding and Exploration for Efficient and Versatile Embodied Navigation

    cs.CV 2025-07 conditional novelty 6.0 of 10

    MTU3D unifies visual grounding and frontier-based exploration in a single transformer, achieving state-of-the-art success rates on HM3D-OVON, GOAT-Bench, SG3D, and A-EQA after large-scale vision-language-exploration p...

  3. Dense360: Dense Understanding from Omnidirectional Panoramas

    cs.CV 2025-06 reject novelty 6.0 of 10

    Introduces a 160K-panorama auto-annotated dataset, a dense captioning and grounding benchmark, and ERP-RoPE; fine-tuning Qwen2.5VL on the data lifts benchmark scores.

Reference graph

Works this paper leans on

87 extracted references · 55 canonical work pages · cited by 3 Pith papers

  1. [1]

    Rocket-1: Master open-world interaction with visual-temporal context prompting

    Shaofei Cai, Zihao Wang, Kewei Lian, Zhancun Mu, Xiaojian Ma, Anji Liu, and Yitao Liang. Rocket-1: Master open-world interaction with visual-temporal context prompting. arXiv preprint arXiv:2410.17856, 2024. 4

    arXiv 2024

  2. [2]

    How far are we to gpt-4v? closing the gap to commercial multimodal models with open-source suites

    Zhe Chen, Weiyun Wang, Hao Tian, Shenglong Ye, Zhangwei Gao, Erfei Cui, Wenwen Tong, Kongzhi Hu, Jiapeng Luo, Zheng Ma, et al. How far are we to gpt-4v? closing the gap to commercial multimodal models with open-source suites. arXiv preprint arXiv:2404.16821, 2024. 6

    arXiv 2024

  3. [3]

    Internvl: Scaling up vision foundation models and aligning for generic visual-linguistic tasks

    Zhe Chen, Jiannan Wu, Wenhai Wang, Weijie Su, Guo Chen, Sen Xing, Muyan Zhong, Qinglong Zhang, Xizhou Zhu, Lewei Lu, et al. Internvl: Scaling up vision foundation models and aligning for generic visual-linguistic tasks. In Proceed- ings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 24185–24198, 2024. 6

    2024

  4. [4]

    Yolo-world: Real-time open- vocabulary object detection

    Tianheng Cheng, Lin Song, Yixiao Ge, Wenyu Liu, Xing- gang Wang, and Ying Shan. Yolo-world: Real-time open- vocabulary object detection. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 16901–16911, 2024. 3, 5, 12

    2024

  5. [5]

    Rescaling egocentric vision: Collection, pipeline and challenges for epic-kitchens-100

    Dima Damen, Hazel Doughty, Giovanni Maria Farinella, An- tonino Furnari, Jian Ma, Evangelos Kazakos, Davide Molti- santi, Jonathan Munro, Toby Perrett, Will Price, and Michael Wray. Rescaling egocentric vision: Collection, pipeline and challenges for epic-kitchens-100. International Journal of Computer Vision (IJCV), 130:33–55, 2022. 1, 8

    2022

  6. [6]

    The llama 3 herd of models

    Abhimanyu Dubey, Abhinav Jauhri, Abhinav Pandey, Ab- hishek Kadian, Ahmad Al-Dahle, Aiesha Letman, Akhil Mathur, Alan Schelten, Amy Yang, Angela Fan, et al. The llama 3 herd of models. arXiv preprint arXiv:2407.21783,

    arXiv

  7. [7]

    Videoagent: A memory-augmented multi- modal agent for video understanding

    Yue Fan, Xiaojian Ma, Rujie Wu, Yuntao Du, Jiaqi Li, Zhi Gao, and Qing Li. Videoagent: A memory-augmented multi- modal agent for video understanding. In European Confer- ence on Computer Vision, pages 75–92. Springer, 2025. 1, 2, 3, 6, 8, 12

    2025

  8. [8]

    Env-qa: A video question answering benchmark for comprehensive understanding of dynamic environments

    Difei Gao, Ruiping Wang, Ziyi Bai, and Xilin Chen. Env-qa: A video question answering benchmark for comprehensive understanding of dynamic environments. In Proceedings of the IEEE/CVF international conference on computer vision, pages 1675–1685, 2021. 2, 6

    2021

Show all 87 references
  1. [9]

    Clova: A closed-loop visual assistant with tool usage and update

    Zhi Gao, Yuntao Du, Xintong Zhang, Xiaojian Ma, Wenjuan Han, Song-Chun Zhu, and Qing Li. Clova: A closed-loop visual assistant with tool usage and update. CVPR, 2023. 2, 8

    2023

  2. [10]

    Mindagent: Emergent gaming interaction

    Ran Gong, Qiuyuan Huang, Xiaojian Ma, Hoi V o, Zane Du- rante, Yusuke Noda, Zilong Zheng, Song-Chun Zhu, Demetri Terzopoulos, Li Fei-Fei, et al. Mindagent: Emergent gaming interaction. arXiv preprint arXiv:2309.09971, 2023. 2, 5

    2023 arXiv

  3. [11]

    Ego4d: Around the world in 3,000 hours of egocentric video

    Kristen Grauman, Andrew Westbury, Eugene Byrne, Zachary Chavis, Antonino Furnari, Rohit Girdhar, Jackson Hamburger, Hao Jiang, Miao Liu, Xingyu Liu, et al. Ego4d: Around the world in 3,000 hours of egocentric video. In Proceedings of the IEEE/CVF Conference on Computer Vision ...

    2022

  4. [12]

    Concept- graphs: Open-vocabulary 3d scene graphs for perception and planning

    Qiao Gu, Ali Kuwajerwala, Sacha Morin, Krishna Murthy Jatavallabhula, Bipasha Sen, Aditya Agarwal, Corban Rivera, William Paul, Kirsty Ellis, Rama Chellappa, et al. Concept- graphs: Open-vocabulary 3d scene graphs for perception and planning. In 2024 IEEE International Confere...

    2024

  5. [13]

    Semantic gaussians: Open-vocabulary scene understanding with 3d gaussian splatting

    Jun Guo, Xiaojian Ma, Yue Fan, Huaping Liu, and Qing Li. Semantic gaussians: Open-vocabulary scene understanding with 3d gaussian splatting. arXiv preprint arXiv:2403.15624,

    arXiv

  6. [14]

    Visual program- ming: Compositional visual reasoning without training

    Tanmay Gupta and Aniruddha Kembhavi. Visual program- ming: Compositional visual reasoning without training. In CVPR, 2023. 2, 8

    2023

  7. [15]

    Temporal alignment networks for long-term video

    Tengda Han, Weidi Xie, and Andrew Zisserman. Temporal alignment networks for long-term video. In CVPR, 2022. 1

    2022

  8. [16]

    Chat-scene: Bridging 3d scene and large language models with object identifiers

    Haifeng Huang, Yilun Chen, Zehan Wang, Rongjie Huang, Runsen Xu, Tai Wang, Luping Liu, Xize Cheng, Yang Zhao, Jiangmiao Pang, et al. Chat-scene: Bridging 3d scene and large language models with object identifiers. In The Thirty- eighth Annual Conference on Neural Information P...

    2024

  9. [17]

    An embodied generalist agent in 3d world

    Jiangyong Huang, Silong Yong, Xiaojian Ma, Xiongkun Linghu, Puhao Li, Yan Wang, Qing Li, Song-Chun Zhu, Baox- iong Jia, and Siyuan Huang. An embodied generalist agent in 3d world. arXiv preprint arXiv:2311.12871, 2023. 1, 3

    2023 arXiv

  10. [18]

    Con- ceptfusion: Open-set multimodal 3d mapping

    Krishna Murthy Jatavallabhula, Alihusein Kuwajerwala, Qiao Gu, Mohd Omama, Tao Chen, Alaa Maalouf, Shuang Li, Ganesh Iyer, Soroush Saryazdi, Nikhil Keetha, et al. Con- ceptfusion: Open-set multimodal 3d mapping. arXiv preprint arXiv:2302.07241, 2023. 1, 8

    2023 arXiv

  11. [19]

    Lemma: A multi-view dataset for le arning m ulti-agent m ulti-task a ctivities

    Baoxiong Jia, Yixin Chen, Siyuan Huang, Yixin Zhu, and Song-chun Zhu. Lemma: A multi-view dataset for le arning m ulti-agent m ulti-task a ctivities. In ECCV, 2020. 1

    2020

  12. [20]

    Egotaskqa: Understanding human tasks in egocentric videos

    Baoxiong Jia, Ting Lei, Song-Chun Zhu, and Siyuan Huang. Egotaskqa: Understanding human tasks in egocentric videos. NeurIPS, 2022. 1

    2022

  13. [21]

    Sceneverse: Scaling 3d vision-language learning for grounded scene un- derstanding

    Baoxiong Jia, Yixin Chen, Huangyue Yu, Yan Wang, Xuesong Niu, Tengyu Liu, Qing Li, and Siyuan Huang. Sceneverse: Scaling 3d vision-language learning for grounded scene un- derstanding. In European Conference on Computer Vision, pages 289–310. Springer, 2025. 3

  14. [22]

    Habitat synthetic scenes dataset (hssd-200): An analysis of 3d scene 9 scale and realism tradeoffs for objectgoal navigation

    Mukul Khanna, Yongsen Mao, Hanxiao Jiang, Sanjay Haresh, Brennan Shacklett, Dhruv Batra, Alexander Clegg, Eric Un- dersander, Angel X Chang, and Manolis Savva. Habitat synthetic scenes dataset (hssd-200): An analysis of 3d scene 9 scale and realism tradeoffs for objectgoal nav...

    2024

  15. [23]

    Text-conditioned resam- pler for long form video understanding

    Bruno Korbar, Yongqin Xian, Alessio Tonioni, Andrew Zis- serman, and Federico Tombari. Text-conditioned resam- pler for long form video understanding. arXiv preprint arXiv:2312.11897, 2023. 1

    2023 arXiv

  16. [24]

    Llama-vid: An im- age is worth 2 tokens in large language models

    Yanwei Li, Chengyao Wang, and Jiaya Jia. Llama-vid: An im- age is worth 2 tokens in large language models. In European Conference on Computer Vision, pages 323–340. Springer,

  17. [25]

    Video-llava: Learning united visual rep- resentation by alignment before projection

    Bin Lin, Yang Ye, Bin Zhu, Jiaxi Cui, Munan Ning, Peng Jin, and Li Yuan. Video-llava: Learning united visual rep- resentation by alignment before projection. arXiv preprint arXiv:2311.10122, 2023. 1, 6

    2023 arXiv

  18. [26]

    Egocentric video-language pretraining

    Kevin Qinghong Lin, Jinpeng Wang, Mattia Soldan, Michael Wray, Rui Yan, Eric Z Xu, Difei Gao, Rong-Cheng Tu, Wen- zhe Zhao, Weijie Kong, et al. Egocentric video-language pretraining. Advances in Neural Information Processing Sys- tems, 35:7575–7586, 2022. 1, 8

    2022

  19. [27]

    Visual instruction tuning

    Haotian Liu, Chunyuan Li, Qingyang Wu, and Yong Jae Lee. Visual instruction tuning. NeurIPS, 2024. 1

    2024

  20. [28]

    Best practices and lessons learned on synthetic data

    Ruibo Liu, Jerry Wei, Fangyu Liu, Chenglei Si, Yanzhe Zhang, Jinmeng Rao, Steven Zheng, Daiyi Peng, Diyi Yang, Denny Zhou, et al. Best practices and lessons learned on synthetic data. In First Conference on Language Modeling,

  21. [29]

    Sqa3d: Situated question answering in 3d scenes

    Xiaojian Ma, Silong Yong, Zilong Zheng, Qing Li, Yitao Liang, Song-Chun Zhu, and Siyuan Huang. Sqa3d: Situated question answering in 3d scenes. In ICLR, 2023. 8

    2023

  22. [30]

    Video-chatgpt: Towards detailed video understanding via large vision and language models

    Muhammad Maaz, Hanoona Rasheed, Salman Khan, and Fahad Shahbaz Khan. Video-chatgpt: Towards detailed video understanding via large vision and language models. arXiv preprint arXiv:2306.05424, 2023. 8

    2023 arXiv

  23. [31]

    Egoloc: Revisiting 3d object localization from egocentric videos with visual queries

    Jinjie Mai, Abdullah Hamdi, Silvio Giancola, Chen Zhao, and Bernard Ghanem. Egoloc: Revisiting 3d object localization from egocentric videos with visual queries. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 45–57, 2023. 2, 5

    2023

  24. [32]

    Openeqa: Embodied question answering in the era of foundation models

    Arjun Majumdar, Anurag Ajay, Xiaohan Zhang, Pranav Putta, Sriram Yenamandra, Mikael Henaff, Sneha Silwal, Paul Mc- vay, Oleksandr Maksymets, Sergio Arnaud, et al. Openeqa: Embodied question answering in the era of foundation models. In Proceedings of the IEEE/CVF Conference on...

    2024

  25. [33]

    Egoschema: A diagnostic benchmark for very long- form video language understanding

    Karttikeya Mangalam, Raiymbek Akshulakov, and Jitendra Malik. Egoschema: A diagnostic benchmark for very long- form video language understanding. NeurIPS, 2024. 1, 8

    2024

  26. [34]

    Howto100m: Learning a text-video embedding by watching hundred mil- lion narrated video clips

    Antoine Miech, Dimitri Zhukov, Jean-Baptiste Alayrac, Makarand Tapaswi, Ivan Laptev, and Josef Sivic. Howto100m: Learning a text-video embedding by watching hundred mil- lion narrated video clips. In ICCV, 2019. 1

    2019

  27. [35]

    Gpt-4 technical report

    OpenAI. Gpt-4 technical report. arXiv preprint arXiv:2303.08774, 2023. 8

    2023 arXiv

  28. [36]

    Dinov2: Learning robust visual features without supervision

    Maxime Oquab, Timothée Darcet, Théo Moutakanni, Huy V o, Marc Szafraniec, Vasil Khalidov, Pierre Fernandez, Daniel Haziza, Francisco Massa, Alaaeldin El-Nouby, et al. Dinov2: Learning robust visual features without supervision. arXiv preprint arXiv:2304.07193, 2023. 12

    2023 arXiv

  29. [37]

    Egovlpv2: Egocentric video-language pre-training with fusion in the backbone

    Shraman Pramanick, Yale Song, Sayan Nag, Kevin Qinghong Lin, Hardik Shah, Mike Zheng Shou, Rama Chellappa, and Pengchuan Zhang. Egovlpv2: Egocentric video-language pre-training with fusion in the backbone. In Proceedings of the IEEE/CVF International Conference on Computer Vis...

    2023

  30. [38]

    Learning transferable visual models from natural language supervision

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. Learning transferable visual models from natural language supervision

  31. [39]

    Sam 2: Segment any- thing in images and videos

    Nikhila Ravi, Valentin Gabeur, Yuan-Ting Hu, Ronghang Hu, Chaitanya Ryali, Tengyu Ma, Haitham Khedr, Roman Rädle, Chloe Rolland, Laura Gustafson, et al. Sam 2: Segment any- thing in images and videos. arXiv preprint arXiv:2408.00714,

    arXiv

  32. [40]

    Concepta- gent: Llm-driven precondition grounding and tree search for robust task planning and execution

    Corban Rivera, Grayson Byrd, William Paul, Tyler Feldman, Meghan Booker, Emma Holmes, David Handelman, Bethany Kemp, Andrew Badger, Aurora Schmidt, et al. Concepta- gent: Llm-driven precondition grounding and tree search for robust task planning and execution. arXiv preprint a...

    arXiv

  33. [41]

    Scene representation transformer: Geometry-free novel view synthe- sis through set-latent scene representations

    Mehdi SM Sajjadi, Henning Meyer, Etienne Pot, Urs Bergmann, Klaus Greff, Noha Radwan, Suhani V ora, Mario Luˇci´c, Daniel Duckworth, Alexey Dosovitskiy, et al. Scene representation transformer: Geometry-free novel view synthe- sis through set-latent scene representations. In P...

    2022

  34. [42]

    Habitat: A platform for embodied ai research

    Manolis Savva, Abhishek Kadian, Oleksandr Maksymets, Yili Zhao, Erik Wijmans, Bhavana Jain, Julian Straub, Jia Liu, Vladlen Koltun, Jitendra Malik, et al. Habitat: A platform for embodied ai research. In ICCV, 2019. 8, 21

    2019

  35. [43]

    Structure-from-motion revisited

    Johannes Lutz Schönberger and Jan-Michael Frahm. Structure-from-motion revisited. In Conference on Computer Vision and Pattern Recognition (CVPR), 2016. 5

    2016

  36. [44]

    Scene representation networks: Continuous 3d-structure- aware neural scene representations

    Vincent Sitzmann, Michael Zollhöfer, and Gordon Wetzstein. Scene representation networks: Continuous 3d-structure- aware neural scene representations. Advances in Neural In- formation Processing Systems, 32, 2019. 1

    2019

  37. [45]

    Vipergpt: Visual inference via python execution for reasoning

    Dídac Surís, Sachit Menon, and Carl V ondrick. Vipergpt: Visual inference via python execution for reasoning. In ICCV,

  38. [46]

    Movieqa: Understanding stories in movies through question-answering

    Makarand Tapaswi, Yukun Zhu, Rainer Stiefelhagen, Anto- nio Torralba, Raquel Urtasun, and Sanja Fidler. Movieqa: Understanding stories in movies through question-answering. In CVPR, 2016. 1

    2016

  39. [47]

    Gemini: a family of highly capable multimodal models

    Gemini Team, Rohan Anil, Sebastian Borgeaud, Yonghui Wu, Jean-Baptiste Alayrac, Jiahui Yu, Radu Soricut, Johan Schalkwyk, Andrew M Dai, Anja Hauth, et al. Gemini: a family of highly capable multimodal models. arXiv preprint arXiv:2312.11805, 2023. 1, 8

    2023 arXiv

  40. [48]

    Gemini 1.5: Unlocking mul- timodal understanding across millions of tokens of context

    Gemini Team, Petko Georgiev, Ving Ian Lei, Ryan Burnell, Libin Bai, Anmol Gulati, Garrett Tanzer, Damien Vincent, 10 Zhufeng Pan, Shibo Wang, et al. Gemini 1.5: Unlocking mul- timodal understanding across millions of tokens of context. arXiv preprint arXiv:2403.05530, 2024. 1, 8

    2024 arXiv

  41. [49]

    Dust3r: Geometric 3d vision made easy

    Shuzhe Wang, Vincent Leroy, Yohann Cabon, Boris Chidlovskii, and Jerome Revaud. Dust3r: Geometric 3d vision made easy. In Proceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recognition, pages 20697–20709, 2024. 5, 15

    2024

  42. [50]

    Lstp: Language-guided spatial-temporal prompt learning for long-form video-text understanding

    Yuxuan Wang, Yueqian Wang, Pengfei Wu, Jianxin Liang, Dongyan Zhao, and Zilong Zheng. Lstp: Language-guided spatial-temporal prompt learning for long-form video-text understanding. arXiv preprint arXiv:2402.16050, 2024. 8

    2024 arXiv

  43. [51]

    Jarvis-1: Open-world multi-task agents with memory-augmented multimodal language models

    Zihao Wang, Shaofei Cai, Anji Liu, Yonggang Jin, Jinbing Hou, Bowei Zhang, Haowei Lin, Zhaofeng He, Zilong Zheng, Yaodong Yang, et al. Jarvis-1: Open-world multi-task agents with memory-augmented multimodal language models. arXiv preprint arXiv:2311.05997, 2023. 1

    2023 arXiv

  44. [52]

    Describe, explain, plan and select: Interactive planning with large language models enables open-world multi-task agents

    Zihao Wang, Shaofei Cai, Anji Liu, Xiaojian Ma, and Yitao Liang. Describe, explain, plan and select: Interactive planning with large language models enables open-world multi-task agents. NeurIPS, 2023

    2023

  45. [53]

    Omnijarvis: Unified vision-language-action tokenization enables open-world instruction following agents

    Zihao Wang, Shaofei Cai, Zhancun Mu, Haowei Lin, Ceyao Zhang, Xuejie Liu, Qing Li, Anji Liu, Xiaojian Ma, and Yitao Liang. Omnijarvis: Unified vision-language-action tokenization enables open-world instruction following agents. arXiv preprint arXiv:2407.00114, 2024. 1

    2024 arXiv

  46. [54]

    Compressed vision for efficient video understanding

    Olivia Wiles, Joao Carreira, Iain Barr, Andrew Zisserman, and Mateusz Malinowski. Compressed vision for efficient video understanding. In ACCV, 2022. 1

    2022

  47. [55]

    Visual chatgpt: Talking, draw- ing and editing with visual foundation models

    Chenfei Wu, Shengming Yin, Weizhen Qi, Xiaodong Wang, Zecheng Tang, and Nan Duan. Visual chatgpt: Talking, draw- ing and editing with visual foundation models. arXiv preprint arXiv:2303.04671, 2023. 2

    2023 arXiv

  48. [56]

    Towards long-form video understanding

    Chao-Yuan Wu and Philipp Krahenbuhl. Towards long-form video understanding. In CVPR, 2021. 1

    2021

  49. [57]

    Next-qa: Next phase of question-answering to explaining temporal actions

    Junbin Xiao, Xindi Shang, Angela Yao, and Tat-Seng Chua. Next-qa: Next phase of question-answering to explaining temporal actions. In CVPR, 2021. 8

    2021

  50. [58]

    Set-of-mark prompting unleashes extraordinary visual grounding in gpt-4v

    Jianwei Yang, Hao Zhang, Feng Li, Xueyan Zou, Chunyuan Li, and Jianfeng Gao. Set-of-mark prompting unleashes extraordinary visual grounding in gpt-4v. arXiv preprint arXiv:2310.11441, 2023. 4

    2023 arXiv

  51. [59]

    Doraemongpt: Toward understanding dynamic scenes with large language models (exemplified as a video agent)

    Zongxin Yang, Guikun Chen, Xiaodi Li, Wenguan Wang, and Yi Yang. Doraemongpt: Toward understanding dynamic scenes with large language models (exemplified as a video agent). 2024. 8

    2024

  52. [60]

    Video-llama: An instruction-tuned audio-visual language model for video un- derstanding

    Hang Zhang, Xin Li, and Lidong Bing. Video-llama: An instruction-tuned audio-visual language model for video un- derstanding. arXiv preprint arXiv:2306.02858, 2023. 8

    2023 arXiv

  53. [61]

    Worldqa: Multimodal world knowledge in videos through long-chain reasoning

    Yuanhan Zhang, Kaichen Zhang, Bo Li, Fanyi Pu, Christo- pher Arif Setiadharma, Jingkang Yang, and Ziwei Liu. Worldqa: Multimodal world knowledge in videos through long-chain reasoning. arXiv preprint arXiv:2405.03272, 2024. 8

    2024 arXiv

  54. [62]

    Mmicl: Empowering vision-language model with multi-modal in-context learning

    Haozhe Zhao, Zefan Cai, Shuzheng Si, Xiaojian Ma, Kaikai An, Liang Chen, Zixuan Liu, Sheng Wang, Wenjuan Han, and Baobao Chang. Mmicl: Empowering vision-language model with multi-modal in-context learning. arXiv preprint arXiv:2309.07915, 2023. 1

    2023 arXiv

  55. [63]

    Learning video representations from large language models

    Yue Zhao, Ishan Misra, Philipp Krähenbühl, and Rohit Gird- har. Learning video representations from large language models. In CVPR, 2023. 14

    2023

  56. [64]

    3d-vista: Pre-trained transformer for 3d vision and text alignment

    Ziyu Zhu, Xiaojian Ma, Yixin Chen, Zhidong Deng, Siyuan Huang, and Qing Li. 3d-vista: Pre-trained transformer for 3d vision and text alignment. In ICCV, 2023. 1, 3

    2023

  57. [65]

    open", "close

    Ziyu Zhu, Zhuofan Zhang, Xiaojian Ma, Xuesong Niu, Yixin Chen, Baoxiong Jia, Zhidong Deng, Siyuan Huang, and Qing Li. Unifying 3d vision-language understanding via prompt- able queries. In European Conference on Computer Vision, pages 188–206. Springer, 2025. 1 11 A. Fields of...

    2025

  58. [66]

    What is the color of the object in the image?

    Use double quotes for all string arguments when using tools, avoiding any single quotes. If the tool has two arguments, format the output within brack- ets, like this: ("What is the color of the object in the image?", 16)

  59. [67]

    Prioritize using retrieve_objects_by_appearance, retrieve_objects_by_environment, and frame_localization for object and frame local- ization, rather than query_database

  60. [68]

    Once the relevant frames are located, use frame_VQA to ask questions about these frames

  61. [69]

    Use frame_VQA to provide these answers

    For "where" questions, interpret them as spatial location queries, not as requests to identify specific frames. Use frame_VQA to provide these answers

  62. [70]

    What color are the blinds?

    If the information is insufficient for a precise response, generate a response based on the available 15 data. Use the following format: Question: the input question you must answer Thought: you should always think about what to do Action: the action to take, should be one of ...

  63. [71]

    The task should involve no more than 2 objects from the room

  64. [72]

    The robot should complete the task using ba- sic actions like GOTO, OPEN, CLOSE, PICK, and PLACE

  65. [73]

    If the robot asks for the location of an object, prompt it to search rather than giving explicit details

  66. [74]

    a dish sponge

    Use general object categories instead of specific IDs (e.g., say "a dish sponge" instead of "dish sponge 1")

  67. [75]

    Once the task is completed, express satisfaction and thank the robot

    Adjust the task if the robot encounters difficulties. Once the task is completed, express satisfaction and thank the robot. Start by initiating a casual conversation and assign- ing a simple task! You are acting as a robot in an apartment. The available receptacles are: {recep...

  68. [76]

    Use the CHAT tool frequently to communicate in a casual manner, keeping the user informed of your progress

  69. [77]

    Example: GOTO(’glass’), then PICK(’glass’); GOTO(’fridge’), then OPEN(’fridge’)

    For every action involving an object or re- ceptacle, first GOTO the target and then perform actions like PICK, PLACE, OPEN, or CLOSE. Example: GOTO(’glass’), then PICK(’glass’); GOTO(’fridge’), then OPEN(’fridge’)

  70. [78]

    Ensure your inventory is empty before picking up a new object

  71. [79]

    Use GOTO, OPEN, and CLOSE to check inside these recepta- cles

    The SEARCH tool can find objects by navigating the room, but it cannot check inside articulated re- ceptacles (like fridges or microwaves). Use GOTO, OPEN, and CLOSE to check inside these recepta- cles

  72. [80]

    Where can I find something to drink

    Before completing the task, use CHAT to confirm the user’s satisfaction. Use the following format: 21 Task: the initial task assigned by the user Thought: you should always think about what to do Action: the action to take, should be one of [{tool_names}] Action Input: the inp...

  73. [81]

    The stone floor is speckled with darker and lighter tones, creating a mosaic-like pattern

    **Natural Stone Walls & Floor:** The walls and floor are made of natural stone, which adds to the rustic aesthetic. The stone floor is speckled with darker and lighter tones, creating a mosaic-like pattern

  74. [82]

    **Wooden Beams:** There are exposed wooden beams supporting the ceiling, giving a sturdy and traditional feel

  75. [83]

    The door on the left is ajar, revealing a room with a wooden table and a lamp stand

    **Wooden Doors:** Two wooden doors are present. The door on the left is ajar, revealing a room with a wooden table and a lamp stand. The door on the right stands tall against the stone wall

  76. [84]

    **Lighting Fixtures:** A charming wall-mounted lamp with a cream shade is visible, providing warm lighting

  77. [85]

    **Interior Details:** There is also a wooden chest with a unique metal clasp, suggesting it’s either a storage piece or perhaps a decorative item

  78. [86]

    **Wooden Cabinets:** The small alcove on the left features wooden cabinets, which implies functional storage space within the room

  79. [87]

    wine glass

    **Entrance/Exit:** There is an open gate leading into the room on the left, furthering the visual appeal of the space, enhancing the sense of depth and accessibility within the interior. The combination of natural materials, simple yet elegant furniture, combined with function...

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.