diveXplore at the Video Browser Showdown 2024

Klaus Schoeffmann; Sahar Nasirihaghighi

REVIEW 3 major objections 5 minor 24 references

diveXplore at the Video Browser Showdown 2024

T0 review · 3 major / 5 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read diveXplore 2024's redesign makes it a competitive video search system for the Video Browser Showdown 2024, its authors argue.

desk verdict A clear, incremental VBS system description whose 'very competitive' claim needs evidence; fine for peer review if the claim is softened. read the letter →

arxiv 2508.20560 v1 pith:4YFXKJPM submitted 2025-08-28 cs.MM

Klaus Schoeffmann , Sahar Nasirihaghighi This is my paper · ORCID

classification cs.MM

keywords videoretrievalinteractivesearchanalysisBrowserShowdownOpenCLIPFAISSkeyframeextractionqueryserver

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 describes diveXplore 2024, an interactive video retrieval system entered in the Video Browser Showdown 2024 competition, and argues that a substantial redesign makes it very competitive. The redesign adds OpenCLIP image/text embeddings trained on LAION-2B for free-text and visual similarity search, a query server that can run several queries in parallel and merge their results, a keyboard-first interface for rapidly inspecting a keyframe's video context, and an exploration view over large clusters of similar videos. A sympathetic reader would care because VBS tests known-item search, ad-hoc video search, and question answering under time pressure, so the paper shows what one team bets on: a generic embedding model plus flexible middleware and interface engineering, with domain-specific analysis reserved for the surgical dataset. The paper does not report competition results; its claim rests on the system's architecture and component choices.

What carries the argument

The load-bearing mechanism is a FAISS index of OpenCLIP ViT-H/14 embeddings computed for every keyframe, which lets the system answer both free-text and visual similarity queries from one representation. Around that sits a Node.js query server, the named central component, that can forward parts of a query to different index instances and to the metadata database, then merge results by filtering, ranking, or temporal constraints, so heterogeneous searches compose in a single interactive loop. Completing the machinery are the interface optimizations — hover-to-inspect neighboring keyframes, keyboard-only navigation, video summaries and shot lists, direct similarity search, and the Xplore view over clusters of similar videos — that speed up the human's inspection and exploration of candidate results.

What would settle it

The published VBS 2024 evaluation results would settle the 'very competitive' claim: if diveXplore finishes at or below the median on known-item search for the Marine or Surgery datasets, the claim is contradicted. Independently, take a sample of text queries from each dataset, retrieve the top-10 keyframes from the FAISS index by embedding similarity, and measure how often a relevant keyframe appears; low recall on Marine or Surgery would show the load-bearing embedding index does not transfer to those domains.

Watch

Extended reading notes

Core claim

On the authors' own terms, diveXplore 2024 is a competitive interactive video retrieval system because every keyframe of every video is embedded once with OpenCLIP ViT-H/14 trained on LAION-2B and stored in a FAISS index; the query server can split a user's request into parallel embedding searches and metadata filters and merge the results, including temporal merges that require ordered matches in nearby shots; and the interface lets users inspect shot lists, play videos, view summaries, jump across keyframes from the same video, and explore clusters of similar videos. The same pipeline is applied to all three VBS datasets (V3C, Marine, Surgery), with shot detection for V3C and Marine, ORB-based keyframe extraction for surgery, and additional analysis of objects, scenes, events, OCR text, and surgical actions. The authors assert that this combination of an improved OpenCLIP embedding index, a flexible distributed query server, and an optimized interface makes the system 'very competitive' at VBS 2024, and they single out the surgical content analysis as essential because laparoscopic footage is highly redundant.

Load-bearing premise

The entire search quality rests on the untested assumption that OpenCLIP embeddings trained on general web images rank the relevant keyframes near the top for the V3C, Marine, and Surgery datasets alike; if that transfer fails, the parallel query server and interface cannot rescue retrieval.

Editorial extensions

If this is right

One embedding index serves both free-text and visual similarity search across all three datasets, so the same interaction works for visual known-item search and textual known-item search.
Parallel query execution lets users combine embedding search with metadata filters, such as object detections or scene concepts, in a single step instead of running separate searches.
Temporal merging over multiple free-text queries supports searches of the form 'first X, then Y nearby', which single-shot ranking cannot express.
The surgical-specific analysis is presented as crucial for handling the highly redundant laparoscopic content, suggesting that generic embeddings alone are not enough for that dataset.
Fast context inspection and cluster exploration are promoted as necessary to keep both expert and novice users quick enough for VBS's timed sessions.

Reading between the lines

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

The paper's competitiveness claim is not tested inside the paper; the actual VBS 2024 standings, which appear in the competition proceedings, would be the direct evidence for whether the redesign paid off.
OpenCLIP was trained on general web image-text pairs, so its transfer to diving and laparoscopic video is far from guaranteed; measuring per-domain retrieval precision, or fine-tuning on domain keyframes, would tell whether the embedding index is the weak link.
The query-server pattern — splitting a query between an embedding index and a metadata store and merging the results — is a generic multimodal retrieval middleware design that could be reused for photo archives, medical image search, or other media collections without the video-specific UI.
The temporal merging capability suggests a useful experiment: compare success rates on event-level known-item queries ('find the clip where X happens, then Y') against single-query text search on V3C, which would quantify the value of the parallel query server beyond speed.

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.

Desk Editor's Note

read the letter

Quick take: diveXplore 2024 is a modest but clearly described update to an existing interactive video retrieval system. The new pieces—OpenCLIP ViT-H/14 embeddings, a parallel query server, and UI tweaks—are genuine changes. The main weakness is that the claim of being 'very competitive' rests on no data at all.

The paper does well at what a VBS system description should do. It gives a readable architecture overview, Table 2 is a clean inventory of which analysis components apply to which dataset, and the UI section explains context inspection, keyboard navigation, and the cluster view without overclaiming. The integration of existing models (TransNetv2, YOLOv7, CRAFT, the surgery-specific analyses) is sensible, and the parallel query server is a real architectural improvement over prior diveXplore versions.

The soft spots are in proportion to the genre. This is not a research paper with an experiment; it's a competition system description. The sentence in Section 3 that the system is 'very competitive' is unsupported. No VBS 2024 scores, no offline retrieval evaluation, no comparison with other systems. The stress-test concern about OpenCLIP transferring from LAION-2B web images to Marine and Surgery video is legitimate, and more acute for the Marine dataset, where OpenCLIP is the only content-based signal. The authors themselves imply that generic embeddings are not enough for surgery ('specific content analysis is crucial'), but they don't validate whether the OpenCLIP index actually ranks relevant keyframes first. That loading is worth fixing even for a demo: either soften the claim to 'we expect it to be competitive' or add a small offline retrieval experiment.

Minor notes: the Marine keyframe subsampling interval is a heuristic but fine; self-citations point to real prior work, so not a concern. No code or data is shipped, which limits reproducibility, but that's typical for this track.

Who should read this: people in the VBS/interactive video retrieval community, and anyone designing a query-server architecture for multimodal retrieval. A serious referee can handle it. It deserves peer review, not a desk reject, but it should be accepted (or at least revised) with the effectiveness claim toned down and, ideally, some evidence or a pointer to public VBS results.

Referee Report

3 major / 5 minor

Summary. The paper describes diveXplore 2024, an interactive video retrieval system for the Video Browser Showdown (VBS) 2024. The system integrates OpenCLIP ViT-H/14 embeddings for free-text and similarity search, a redesigned query server that can distribute and merge parallel queries, and an optimized user interface with new exploration views. The authors state that these changes make the system 'very competitive.' The paper presents the system architecture, the analysis components summarized in Table 2, and several UI features, but it contains no experimental evaluation and no VBS 2024 results.

Significance. If the system performs as claimed, this paper provides a useful system description for the VBS community, and the combination of OpenCLIP embeddings with a parallel query server is a coherent architectural choice. The manuscript is clearly structured and honest about the components used. However, its significance is limited by the absence of any evaluation: the central claim of competitiveness is not backed by retrieval experiments, VBS scores, or ablations. As a system description it is informative, but as a demonstration of effectiveness it currently falls short.

major comments (3)

[3. Conclusion] The claim that the improved index, distributed query server, and UI optimizations 'make our system very competitive' is not supported by any experimental evidence in the manuscript. No VBS 2024 results, offline retrieval evaluations, or ablations are reported. Since competitiveness is the paper's central assertion, this is a load-bearing omission. The authors should either add evaluation data (e.g., VBS 2024 scores or an offline known-item search experiment on the three datasets) or substantially temper the claim to a purely descriptive system paper.
[2.1, Table 2] For the Marine dataset, OpenCLIP embeddings are the only content-based retrieval signal; Table 2 shows no other analysis applied to M. The paper provides no validation that a model trained on LAION-2B transfers to underwater diving videos. For the Surgery dataset, the authors themselves state in Section 3 that 'specific content analysis' is 'crucial' because of highly redundant content, implicitly acknowledging that generic embeddings are insufficient, yet no experiment demonstrates that OpenCLIP ranks relevant surgical or marine keyframes above irrelevant ones. A targeted retrieval evaluation on these two domain-specific collections is needed to substantiate the competitiveness claim.
[2.2] The query server is described as enabling 'flexible and efficient' parallel queries, but no performance measurements, scaling tests, or comparisons with the previous diveXplore version are provided. Without timing data or a relative comparison, the efficiency claim remains unsupported. Adding at least a simple latency benchmark or a qualitative comparison of query merging strategies would help.

minor comments (5)

[2.1, Table 2] The table cites 'OpenCLIP ViT-H/14 [12]', but reference [12] is Radford et al.'s CLIP paper, not the OpenCLIP implementation; the correct citation for OpenCLIP is reference [4].
[1] The phrase 'Ad-hoch search' should be 'ad-hoc search', and 'datset' should be 'dataset'.
[2.1] The sentence 'Figure 1 shows the that architecture' contains an extra word; 'the that' should be 'the'.
[Abstract] The word 'redesigend' should be 'redesigned'.
[1] In the example query description, 'baloon' should be 'balloon'.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the paper is a system description with no derivation that reduces to its inputs.

full rationale

The paper is a short interactive-video-retrieval system description. It contains no equations, no fitted parameters, and no formal derivation chain whose conclusion is assumed in its premises. The components (OpenCLIP ViT-H/14, FAISS, TransNetv2, YOLOv7, etc.) are cited external models and datasets, not results derived within the paper. The paper's central claim that diveXplore is "very competitive" (Section 3) is an assertion based on integrating those components, not a quantity computed from them, so it cannot be circular. The self-citations ([11], [16], [17], [18]) point to prior toolchain components such as endoscopic keyframe extraction and action recognition; these are supporting references for specific analysis modules, not used to justify the paper's own competitiveness claim. The skeptical concern that OpenCLIP's transfer to Marine and Surgery domains is unvalidated is a legitimate correctness/evidence gap, but it is not a circularity issue. Therefore the appropriate finding is no significant circularity.

Assumptions & free parameters 1 free parameters · 3 assumptions · 0 invented entities

The system is assembled from existing models and tools; no new entities are invented. The central claim of competitiveness rests on unvalidated assumptions about the transfer of pre-trained models to the VBS datasets, and the reported sampling interval is a hand-chosen design parameter without ablation.

free parameters (1)

Marine keyframe sampling interval = 1 second
Chosen by hand for uniform temporal subsampling of marine videos; no ablation or justification is provided, and retrieval performance may depend on it.

assumptions (3)

domain assumption OpenCLIP ViT-H/14 (LAION-2B) embeddings capture semantic similarity relevant to VBS retrieval queries.
The FAISS similarity and free-text search rely on this transfer; no evidence is given for V3C, Marine, or Surgery collections.
domain assumption Shot boundary detection via TransNetv2 and keyframe extraction produce representative frames for retrieval.
All subsequent analysis and search operate on these keyframes; accuracy is assumed without evaluation.
domain assumption Off-the-shelf detectors for objects, scenes, text, and actions provide correct metadata for combined queries.
Metadata filters and merged queries depend on these labels; no dataset-specific validation is reported.

how reviews work

0 comments

Cite this review

Pith. "Pith review of diveXplore at the Video Browser Showdown 2024." pith.science (2026). https://pith.science/paper/4YFXKJPM

@misc{pith2026250820560,
  author       = {Pith},
  title        = {Pith review of: diveXplore at the Video Browser Showdown 2024},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/4YFXKJPM}},
  note         = {Machine review of arXiv:2508.20560}
}

read the original abstract

According to our experience from VBS2023 and the feedback from the IVR4B special session at CBMI2023, we have largely revised the diveXplore system for VBS2024. It now integrates OpenCLIP trained on the LAION-2B dataset for image/text embeddings that are used for free-text and visual similarity search, a query server that is able to distribute different queries and merge the results, a user interface optimized for fast browsing, as well as an exploration view for large clusters of similar videos (e.g., weddings, paraglider events, snow and ice scenery, etc.).

Figures

Figures reproduced from arXiv: 2508.20560 by the authors.

**Figure 1.** diveXplore 2024 architecture 2.1 Video Analysis The first step of the analysis is shot segmentation and keyframe extraction, which is done with TransNetv2 [20] for V3C and a specific keyframe extraction algorithm for endoscopic videos [17] that is based on ORB keypoint tracking. As keyframes, the middle of the shot is used. For videos in the marine dataset, we simply use uniform temporal subsampling, with a 1-second… view at source ↗

**Figure 2.** diveXplore User Interface position moves; this should allow the user to very quickly inspect the video context of the keyframe. The result list is strongly optimized for keyboard-only use. The space key can used to open and close the video summary for a keyframe, the arrow keys are used to navigate between results and pages. The search bar supports an expert mode that can combine several different search modalities… view at source ↗

**Figure 3.** Different views of diveXplore [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗

Review history (2 revisions) →

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

24 extracted references · 11 canonical work pages

[1]

In: Proceedings of the IEEE/CVF international conference on computer vision

Baek, J., Kim, G., Lee, J., Park, S., Han, D., Yun, S., Oh, S.J., Lee, H.: What is wrong with scene text recognition model comparisons? dataset and model analysis. In: Proceedings of the IEEE/CVF international conference on computer vision. pp. 4715–4723 (2019)

work page 2019
[2]

In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

Baek, Y., Lee, B., Han, D., Yun, S., Lee, H.: Character region awareness for text detection. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 9365–9374 (2019)

work page 2019
[3]

In: Proc

Berns, F., Rossetto, L., Schoeffmann, K., Beecks, C., Awad, G.: V3c1 dataset: An evaluation of content characteristics. In: Proc. of the 2019 on Intl. Conf. on Multimedia Retrieval. pp. 334–338. ACM (2019)

2019
[4]

https://doi.org/10.48550/ARXIV.2212.07143, https://arxiv.org/abs/2212.07143

Cherti, M., Beaumont, R., Wightman, R., Wortsman, M., Ilharco, G., Gordon, C., Schuhmann, C., Schmidt, L., Jitsev, J.: Reproducible scaling laws for contrastive language-image learning (2022). https://doi.org/10.48550/ARXIV.2212.07143, https://arxiv.org/abs/2212.07143

arXiv doi:10.48550/arxiv.2212.07143 2022
[5]

IEEE Transactions on Big Data7(3), 535–547 (2019)

Johnson, J., Douze, M., Jégou, H.: Billion-scale similarity search with GPUs. IEEE Transactions on Big Data7(3), 535–547 (2019)

2019
[6]

In: MultiMedia Modeling: 26th International Conference, MMM 2020, Daejeon, South Korea, Jan- uary 5–8, 2020, Proceedings, Part II 26

Kletz, S., Schoeffmann, K., Leibetseder, A., Benois-Pineau, J., Husslein, H.: In- strument recognition in laparoscopy for technical skill assessment. In: MultiMedia Modeling: 26th International Conference, MMM 2020, Daejeon, South Korea, Jan- uary 5–8, 2020, Proceedings, Part II 26. pp. 589–600. Springer (2020)

work page 2020
[7]

In: International Conference on Multimedia Modeling

Leibetseder, A., Schoeffmann, K.: divexplore 6.0: Itec’s interactive video explo- ration system at vbs 2022. In: International Conference on Multimedia Modeling. pp. 569–574. Springer (2022)

work page 2022
[8]

Multimedia Systems pp

Lokoč, J., Andreadis, S., Bailer, W., Duane, A., Gurrin, C., Ma, Z., Messina, N., Nguyen, T.N., Peška, L., Rossetto, L., et al.: Interactive video retrieval in the age of effective joint embedding deep models: lessons from the 11th vbs. Multimedia Systems pp. 1–24 (2023)

work page 2023

Show all 24 references

[9]

ACM Trans

Lokoč, J., Kovalčík, G., Münzer, B., Schöffmann, K., Bailer, W., Gasser, R., Vrochidis, S., Nguyen, P.A., Rujikietgumjorn, S., Barthel, K.U.: Interactive search or sequential browsing? a detailed analysis of the video browser showdown 2018. ACM Trans. Multimedia Comput. Commun...

2018 doi
[10]

IEEE Trans- actions on Pattern Analysis and Machine Intelligence 42(2), 502–508 (2020), https://doi.org/10.1109/TPAMI.2019.2901464

Monfort, M., Vondrick, C., Oliva, A., Andonian, A., Zhou, B., Ramakrish- nan, K., Bargal, S.A., Yan, T., Brown, L.M., Fan, Q., Gutfreund, D.: Mo- ments in time dataset: One million videos for event understanding. IEEE Trans- actions on Pattern Analysis and Machine Intelligence...

2020
[11]

In: 2023 IEEE 36th International Symposium on Computer-Based Medical Systems (CBMS)

Nasirihaghighi, S., Ghamsarian, N., Stefanics, D., Schoeffmann, K., Husslein, H.: Action recognition in video recordings from gynecologic laparoscopy. In: 2023 IEEE 36th International Symposium on Computer-Based Medical Systems (CBMS). pp. 29–34. IEEE (2023)

2023
[12]

In: International Conference on Machine Learning

Radford, A., Kim, J.W., Hallacy, C., Ramesh, A., Goh, G., Agarwal, S., Sastry, G., Askell, A., Mishkin, P., Clark, J., et al.: Learning transferable visual models from natural language supervision. In: International Conference on Machine Learning. pp. 8748–8763. PMLR (2021)

2021
[13]

Schoeffmann et al

Rossetto, L., Gasser, R., Lokoč, J., Bailer, W., Schoeffmann, K., Muen- zer, B., Souček, T., Nguyen, P.A., Bolettieri, P., Leibetseder, A., Vrochidis, 8 K. Schoeffmann et al. S.: Interactive video retrieval in the age of deep learning – detailed eval- uation of vbs 2019. IEEE ...

2021
[14]

In: MultiMedia Modeling: 27th International Conference, MMM 2021, Prague, Czech Republic, June 22–24, 2021, Proceedings, Part II 27

Rossetto,L.,Gasser,R.,Sauter,L.,Bernstein,A.,Schuldt,H.:Asystemforinterac- tive multimedia retrieval evaluations. In: MultiMedia Modeling: 27th International Conference, MMM 2021, Prague, Czech Republic, June 22–24, 2021, Proceedings, Part II 27. pp. 385–390. Springer (2021)

2021
[15]

arXiv preprint arXiv:2105.01475 (2021)

Rossetto, L., Schoeffmann, K., Bernstein, A.: Insights on the v3c2 dataset. arXiv preprint arXiv:2105.01475 (2021)

2021 arXiv
[16]

In: Proceedings of the 20th International Conference on Content-based Multimedia Indexing (CBMI 2023)

Schoeffmann, K.: divexb: An interactive video retrieval system for beginners. In: Proceedings of the 20th International Conference on Content-based Multimedia Indexing (CBMI 2023). pp. 1–6. IEEE (2023)

2023
[17]

Multimedia Tools and Applications74, 11187–11206 (2015)

Schoeffmann, K., Del Fabro, M., Szkaliczki, T., Böszörmenyi, L., Keckstein, J.: Keyframe extraction in endoscopic video. Multimedia Tools and Applications74, 11187–11206 (2015)

2015
[18]

In: International Conference on Multimedia Modeling

Schoeffmann, K., Stefanics, D., Leibetseder, A.: divexplore at the video browser showdown 2023. In: International Conference on Multimedia Modeling. pp. 684–

2023
[19]

arXiv preprint arXiv:2111.02114 (2021)

Schuhmann, C., Vencu, R., Beaumont, R., Kaczmarczyk, R., Mullis, C., Katta, A., Coombes,T.,Jitsev,J.,Komatsuzaki,A.:Laion-400m:Opendatasetofclip-filtered 400 million image-text pairs. arXiv preprint arXiv:2111.02114 (2021)

2021 arXiv
[20]

arXiv preprint arXiv:2008.04838 (2020)

Souček, T., Lokoč, J.: Transnet v2: an effective deep network architecture for fast shot transition detection. arXiv preprint arXiv:2008.04838 (2020)

2020 arXiv
[21]

In: International conference on machine learning

Tan, M., Le, Q.: Efficientnet: Rethinking model scaling for convolutional neural networks. In: International conference on machine learning. pp. 6105–6114. PMLR (2019)

2019
[22]

In: International Conference on Multimedia Modeling

Truong, Q.T., Vu, T.A., Ha, T.S., Lokoč, J., Wong, Y.H., Joneja, A., Yeung, S.K.: Marine video kit: a new marine video dataset for content-based analysis and retrieval. In: International Conference on Multimedia Modeling. pp. 539–550. Springer (2023)

2023
[23]

arXiv preprint arXiv:2207.02696 (2022)

Wang, C.Y., Bochkovskiy, A., Liao, H.Y.M.: Yolov7: Trainable bag-of-freebies sets new state-of-the-art for real-time object detectors. arXiv preprint arXiv:2207.02696 (2022)

2022 arXiv
[24]

IEEE Transactions on Pattern Analysis and Machine Intelligence 40(6), 1452–1464 (2018), https://doi.org/10.1109/TPAMI.2017.2723009

Zhou, B., Lapedriza, A., Khosla, A., Oliva, A., Torralba, A.: Places: A 10 million image database for scene recognition. IEEE Transactions on Pattern Analysis and Machine Intelligence 40(6), 1452–1464 (2018), https://doi.org/10.1109/TPAMI.2017.2723009

2018

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.

[1] [1]

In: Proceedings of the IEEE/CVF international conference on computer vision

Baek, J., Kim, G., Lee, J., Park, S., Han, D., Yun, S., Oh, S.J., Lee, H.: What is wrong with scene text recognition model comparisons? dataset and model analysis. In: Proceedings of the IEEE/CVF international conference on computer vision. pp. 4715–4723 (2019)

work page 2019

[2] [2]

In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

Baek, Y., Lee, B., Han, D., Yun, S., Lee, H.: Character region awareness for text detection. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 9365–9374 (2019)

work page 2019

[3] [3]

In: Proc

Berns, F., Rossetto, L., Schoeffmann, K., Beecks, C., Awad, G.: V3c1 dataset: An evaluation of content characteristics. In: Proc. of the 2019 on Intl. Conf. on Multimedia Retrieval. pp. 334–338. ACM (2019)

2019

[4] [4]

https://doi.org/10.48550/ARXIV.2212.07143, https://arxiv.org/abs/2212.07143

Cherti, M., Beaumont, R., Wightman, R., Wortsman, M., Ilharco, G., Gordon, C., Schuhmann, C., Schmidt, L., Jitsev, J.: Reproducible scaling laws for contrastive language-image learning (2022). https://doi.org/10.48550/ARXIV.2212.07143, https://arxiv.org/abs/2212.07143

arXiv doi:10.48550/arxiv.2212.07143 2022

[5] [5]

IEEE Transactions on Big Data7(3), 535–547 (2019)

Johnson, J., Douze, M., Jégou, H.: Billion-scale similarity search with GPUs. IEEE Transactions on Big Data7(3), 535–547 (2019)

2019

[6] [6]

In: MultiMedia Modeling: 26th International Conference, MMM 2020, Daejeon, South Korea, Jan- uary 5–8, 2020, Proceedings, Part II 26

Kletz, S., Schoeffmann, K., Leibetseder, A., Benois-Pineau, J., Husslein, H.: In- strument recognition in laparoscopy for technical skill assessment. In: MultiMedia Modeling: 26th International Conference, MMM 2020, Daejeon, South Korea, Jan- uary 5–8, 2020, Proceedings, Part II 26. pp. 589–600. Springer (2020)

work page 2020

[7] [7]

In: International Conference on Multimedia Modeling

Leibetseder, A., Schoeffmann, K.: divexplore 6.0: Itec’s interactive video explo- ration system at vbs 2022. In: International Conference on Multimedia Modeling. pp. 569–574. Springer (2022)

work page 2022

[8] [8]

Multimedia Systems pp

Lokoč, J., Andreadis, S., Bailer, W., Duane, A., Gurrin, C., Ma, Z., Messina, N., Nguyen, T.N., Peška, L., Rossetto, L., et al.: Interactive video retrieval in the age of effective joint embedding deep models: lessons from the 11th vbs. Multimedia Systems pp. 1–24 (2023)

work page 2023