Conceptualizing Multi-scale Wavelet Attention and Ray-based Encoding for Human-Object Interaction Detection

Junn Yong Loo; KokSheik Wong; Quan Bi Pay; Simon See; Vishnu Monn Baskaran

REVIEW 2 major objections 5 minor 45 references

Conceptualizing Multi-scale Wavelet Attention and Ray-based Encoding for Human-Object Interaction Detection

T0 review · 2 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read This paper claims that a wavelet attention backbone combined with ray-based encoding produces an efficient human-object interaction detector that is leaner and faster than a heavier baseline, at a measured accuracy cost.

desk verdict A genuinely new ray-based attention mechanism with an honest write-up, but the headline gain on HICO-DET rests on an ablation that is not controlled. read the letter →

arxiv 2507.10977 v1 pith:5ZPZ6KV2 submitted 2025-07-15 cs.CV cs.AI

Quan Bi Pay , Vishnu Monn Baskaran , Junn Yong Loo , KokSheik Wong , Simon See This is my paper

classification cs.CVcs.AI

keywords human-objectinteractiondetectionwaveletbackboneray-basedencodingmulti-scaleattentionmiddle-orderinteractionsHICO-DETImageNetclassificationefficient

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

This paper tries to establish that a backbone built around multi-scale wavelet-like filtering, combined with a ray-based encoding module, can make human-object interaction detection more efficient without collapsing accuracy. The authors claim the wavelet attention stage captures "middle-order" interactions between low- and high-order features, which patch-based CNNs tend to miss, and the ray encoder gives the decoder a learnable spatial prior that focuses queries on relevant regions. On HICO-DET the full model reaches 24.07 mAP with 34.46M parameters at 10 FPS, versus 29.81 mAP, 55.42M parameters and 8 FPS for the FGAHOI baseline it builds on. The message is that accuracy-efficiency trade-offs in HOI detection can come from the backbone and attention prior, not only from heavier decoders.

What carries the argument

The two load-bearing components are the wavelet attention backbone and the ray-based encoder. The backbone decomposes feature maps with small and large convolutional filters to produce low- and high-frequency bands, concatenates the four directional sub-bands, then uses a pointwise-convolution modulation computed from these bands to refine features; this is how it aggregates middle-order interactions. The ray encoder treats learnable origin points as attention priors: it computes each pixel's distance to every origin, applies a Gaussian point-spread function and an exponential decay to obtain an attenuation map, passes it through softmax, and uses it to modulate features in the frequency domain via global circular convolution. The learned origins converge toward image centers during training, which the paper interprets as the mechanism learning a center-bias prior.

What would settle it

Train the same wavelet backbone plus ray encoder on an object-centric dataset whose subjects are not centered, or on HICO-DET after randomly translating objects away from the center; if the mAP gain from rays disappears or reverses, the claimed mechanism is center-bias exploitation rather than general multi-scale attention.

Watch

Extended reading notes

Core claim

The central claim is that aggregating low- and high-frequency convolutional responses into an attention-like modulation recovers middle-order pixel interactions, and that encoding multi-scale attention through learnable ray origins gives a compact way to emphasize task-relevant regions. The paper reports that adding three ray layers to the wavelet backbone raises ImageNet top-1 accuracy from 73.36% to 74.54% and raises HICO-DET default Full mAP from 20.97% to 24.07%, while using fewer parameters and running faster than the FGAHOI detector that supplies the decoder.

Load-bearing premise

The load-bearing premise is that the ray module's benefit comes from a reusable attention prior, not from exploiting the fact that objects tend to sit in the center of the training images.

Editorial extensions

If this is right

If the central claim holds, HOI detectors can be made leaner by replacing heavy encoder stacks with a ray-based attention prior plus a wavelet backbone, without giving up the DETR-style decoder.
Ray encoding transfers from image classification to HOI detection: the same module that improves ImageNet top-1 accuracy also adds 3.10 mAP on HICO-DET default Full.
The wavelet-only variant already reaches 20.97 mAP on HICO-DET, indicating the backbone itself contributes a large share of the final accuracy, with rays contributing the remaining 3.10 points.
Fewer parameters and higher FPS mean the proposed architecture is better suited than FGAHOI to real-time or resource-constrained HOI applications, at a measured 5.74 mAP accuracy cost.

Reading between the lines

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

The ray module's convergence toward image centers suggests it is essentially a learned spatial prior; a natural extension the paper does not test is applying it to datasets where objects are not centered, and if transfer fails there, the mechanism is dataset-specific rather than a general attention prior.
The ray concept could be ported to other dense-prediction tasks such as object detection or segmentation, where a learnable focal region prior might reduce encoder depth while preserving accuracy.
Parameterizing rays with direction or angle, which the paper lists as future work, would let the model represent off-center or elongated attention regions and might close the gap to FGAHOI without adding decoder parameters.

Share X Bluesky LinkedIn Reddit HN

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

Colleague,

Here is the take: the paper introduces a ray-origin attenuation encoder for HOI detection, which is genuinely new in 2D attention, and it is written honestly. But the central quantitative claim—that adding ray layers to the wavelet backbone gives +3.10 mAP on HICO-DET—is not backed by a controlled ablation. The two models differ in encoder composition and parameter count, so the gain cannot be attributed cleanly to the ray mechanism.

What is new: the ray mechanism models K and Q using learnable 2D anchor points, PSF spread, and exponential decay, producing a Fourier-domain attention map. That use of ray concepts in HOI attention is absent from the cited prior work, and the design is clearly explained. The wavelet backbone is a variant of existing wavelet CNNs, but the aggregation of low/high-order interactions into middle-order ones is a reasonable framing. Internal ablations on ImageNet are consistent: each added ray layer improves top-1 accuracy, and the paper openly reports the trade-off with FGAHOI (24.07 vs 29.81 mAP) and the efficiency gain (34.46M params, 10 FPS). Section V candidly lists limitations, including the data sensitivity of the ray layer.

The soft spots, in proportion. The big one is the stress-test concern: Table III's ray gain is confounded. The implementation details say the final model uses "3 ray-based encoding layers and 3 transformer encoder layers," while the Wavelet-only model presumably keeps six transformer layers; that also explains the parameter difference (36.94 vs 34.46M) and the FPS difference. So the +3.10 mAP could come from the removed transformer layers, the changed parameter budget, or the ray mechanism itself. A matched-composition control—same number and width of layers, swapping only the mechanism—is the minimal fix. Second, there are no external baselines on ImageNet and no multiple seeds with variance, so we cannot tell if the improvements are stable. Third, the center-bias limitation is real but secondary; a matched ablation would still show whether the ray mechanism helps on HICO-DET, even if the benefit is a center prior. Minor points: alpha, beta, and PSF initialization are unspecified, and the code repo should have a commit hash for reproducibility.

Who this is for: researchers working on efficient HOI detectors and novel attention priors. The middle-order interaction story is more asserted than measured, but the ray idea is worth examining.

Recommendation: send to peer review. A serious referee should require the matched-composition ablation and variance reporting before the claims can be accepted. If the ray gain survives that, it is a decent contribution; if not, it is a promising idea that currently overstates its evidence.

Referee Report

2 major / 5 minor

Summary. The paper proposes a wavelet attention-like backbone and a ray-based encoder for human-object interaction (HOI) detection. The wavelet backbone is designed to capture middle-order pixel interactions by aggregating features from filters of different sizes, while the ray-based encoder uses learnable ray origins to produce attenuation maps that guide the decoder's attention. The authors evaluate the model on ImageNet-1K for image classification and on HICO-DET for HOI detection. They report that adding three ray layers improves ImageNet top-1 accuracy from 73.36% to 74.54% and improves HICO-DET default mAP from 20.97% to 24.07%, while using fewer parameters and higher FPS than the FGAHOI baseline.

Significance. If the reported gains are attributable to the proposed mechanisms, the paper makes a modest but useful contribution: it demonstrates a parameter-lean alternative architecture for HOI detection and provides one of the first applications of ray-inspired attention to 2D vision. The manuscript is strengthened by the public release of code, by the monotonic improvements when ray layers are added in the ImageNet ablation (Table II), and by a candid discussion of the ray mechanism's data sensitivity in Section V. However, the central HOI ablation is confounded by differences in model composition, and the paper's own analyses show that the ray mechanism encodes a center bias, so the current evidence does not yet support the broader claims of a general multi-scale attention mechanism.

major comments (2)

[IV.C.1, Table III] The claimed +3.10 mAP gain from ray-based encoding is not supported as a controlled ablation. Table III reports Wavelet with 36.94M parameters and Wavelet+3 Rays with 34.46M parameters, yet the ImageNet ablation in Table II shows that adding three ray layers increases parameters from 9.58M to 10.38M. The most consistent explanation is that the two HOI models differ in encoder composition, not merely in the presence of ray layers; Section IV.A.2 states that the encoder uses '3 ray-based encoding layers and 3 transformer encoder layers.' To attribute the +3.10 mAP to the ray mechanism, the authors must compare models that differ only in the ray layers, with the backbone, the number of transformer layers, and the parameter budget matched. Without such a control, the gain could be caused by the removed transformer layers or by the changed capacity.
[III.B.2, IV.B.2, V.1] The paper's own evidence indicates that the ray mechanism exploits a center bias rather than providing a general attention prior. Ray origins are initialized on a unit circle because 'most of the objects are centered in the image' (Fig. 5(d)), trained origins converge to the image center (Fig. 5(a-c)), and Section V.1 concedes that the ray layer 'is sensitive to the data' and currently has no directionality. The ImageNet and HICO-DET gains may therefore reflect dataset statistics rather than a reusable spatial-reasoning mechanism. The authors should provide a test on a dataset without a strong center bias, or an explicit spatial-jitter experiment, to substantiate transferability; alternatively, the claims should be narrowed to describe a center-prior module.

minor comments (5)

[Table II] Wavelet+1 Ray and Wavelet+2 Rays are both listed with 9.98M parameters, which is surprising because adding a ray layer should change the parameter count; please clarify whether the values are rounded or whether the second ray layer shares parameters with the first.
[IV.B.1] The text calls the ImageNet result 'competitive,' but Table II contains no external baseline such as ResNet, ConvNeXt, or Swin at a comparable parameter and FLOP budget; a direct comparison is needed to support that descriptor.
[Equation (6)] The softmax in Eq. (6) is not fully specified: it is unclear whether normalization is performed over the n ray origins, over the H×W pixel positions, or over both; please state the normalization axis explicitly.
[III.B.2] The architecture description mentions 'representation learning using 2D Fast Fourier transformation' and 'learning in frequency domain,' but no FFT-based operation appears in the equations or figure; please provide the exact frequency-domain computation or remove the claim.
[Fig. 5(d)] The object-center distribution is shown only for ImageNet, yet the ray module is also used on HICO-DET; providing a similar distribution for HICO-DET would help justify the transfer of the center prior.

Circularity Check

0 steps flagged · score 2.0 of 10

No circular derivation; the paper's benchmark numbers are direct test-set measurements and the ray mechanism is a learned module, so the central claims do not reduce to their inputs.

full rationale

We examined the derivation chain from the wavelet backbone equations (Eq. 2-3) through the ray attenuation map (Eq. 6) to the ImageNet and HICO-DET evaluations. The wavelet feature maps are computed from convolutional decompositions, and the ray map is a trainable distance-based attention prior; neither quantity is defined in terms of the final mAP numbers. The ImageNet and HICO-DET results are external test-set measurements, and the ray origins are optimized by backpropagation rather than fitted to the evaluation metric. The only self-citation (ERNet, ref. [16]) appears in a related-works categorization of query-enhancement methods and is not load-bearing for the proposed architecture. The center-bias motivation in Sec. III.B.2 ('most of the objects are centered in the image') is a dataset statistics prior, not a circular reuse of the paper's own output; the later observation that ray sources converge to the center is a training outcome, and the paper explicitly flags data sensitivity in Sec. V.1. The Table III comparison between Wavelet (36.94M params) and Wavelet+3 Rays (34.46M params) is not a matched-composition ablation, so the +3.10 mAP may be partly due to changed encoder composition; this is a correctness/experimental-design concern, not a circularity. No step in the paper equates a prediction with an input by construction, imports a uniqueness theorem, or renames a known result as a derivation.

Assumptions & free parameters 4 free parameters · 4 assumptions · 1 invented entities

The architecture's contribution reduces to two mechanism assumptions: that frequency-band aggregation creates the advertised middle-order interactions, and that learnable distance-based attention from ray origins is a generalizable prior. The ray features (origins, PSF variance, alpha, beta) are learned or hand-set within the paper rather than derived from independent constraints. No new physical entities are introduced; the ray origin is a model parameter, not a measured quantity.

free parameters (4)

ray origin coordinates O (12x2) = learned during training; converge near image center (Fig. 5)
Twelve learnable anchor points parameterize the ray attenuation map in Eq. (5); their converged positions track the ImageNet center-bias and drive the reported gains.
PSF variance = learnable, value not reported
The point spread function in Eq. (6) is a 2D Gaussian with learnable variance; its initial value and range are not stated.
attenuation coefficients alpha, beta = unspecified
Eq. (6) uses beta * exp(-alpha D) to form the query; the paper does not state whether these are learned or fixed, nor their values.
ray origin count n=12 and ray-layer count = 12 origins; 1+1 ray layers in backbone, 3 in encoder
Hand-chosen design choices; the ablation shows accuracy rises with ray layers (Table II) while FPS falls, so the gains depend on these choices.

assumptions (4)

domain assumption Convolution with differently sized filters separates low-frequency from high-frequency information, so concatenating bands yields middle-order interactions (Section III.A.1-2).
The paper asserts that 'a straightforward way to model intermediate interactions within CNN layers is to use filters of varying sizes'; this heuristic borrows the interaction-order concept from [22] but is never measured directly.
domain assumption Multi-resolution decomposition via four frequency bands (Eq. 2) preserves the discriminative information needed for HOI without redundancy.
The backbone's power is assumed to come from the wavelet-style split; no ablation replaces the split with a plain multi-branch convolution of matched cost.
standard math The element-wise multiplication in frequency domain equals depthwise global circular convolution in the spatial domain (cited to [41]), justifying Fourier-domain ray modulation.
Standard signal-processing identity; cited, not derived.
ad hoc to paper Objects in training images are centered, so unit-circle initialization of ray origins provides a useful prior that transfers to HICO-DET.
Section III.B.2 justifies the design by Fig. 5(d); the paper itself flags the resulting data sensitivity in Section V.1.

invented entities (1)

Learnable ray origin (2D anchor point)
purpose: Serves as a pivotal point defining regions of interest; pixels near the origin receive higher attention weight via Eq. (6).
A new model component with no falsifiable handle outside the paper's own benchmark numbers; its behavior (convergence to image center) is shown only on the paper's training distributions.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Conceptualizing Multi-scale Wavelet Attention and Ray-based Encoding for Human-Object Interaction Detection." pith.science (2026). https://pith.science/paper/5ZPZ6KV2

@misc{pith2026250710977,
  author       = {Pith},
  title        = {Pith review of: Conceptualizing Multi-scale Wavelet Attention and Ray-based Encoding for Human-Object Interaction Detection},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/5ZPZ6KV2}},
  note         = {Machine review of arXiv:2507.10977}
}

read the original abstract

Human-object interaction (HOI) detection is essential for accurately localizing and characterizing interactions between humans and objects, providing a comprehensive understanding of complex visual scenes across various domains. However, existing HOI detectors often struggle to deliver reliable predictions efficiently, relying on resource-intensive training methods and inefficient architectures. To address these challenges, we conceptualize a wavelet attention-like backbone and a novel ray-based encoder architecture tailored for HOI detection. Our wavelet backbone addresses the limitations of expressing middle-order interactions by aggregating discriminative features from the low- and high-order interactions extracted from diverse convolutional filters. Concurrently, the ray-based encoder facilitates multi-scale attention by optimizing the focus of the decoder on relevant regions of interest and mitigating computational overhead. As a result of harnessing the attenuated intensity of learnable ray origins, our decoder aligns query embeddings with emphasized regions of interest for accurate predictions. Experimental results on benchmark datasets, including ImageNet and HICO-DET, showcase the potential of our proposed architecture. The code is publicly available at [https://github.com/henry-pay/RayEncoder].

Figures

Figures reproduced from arXiv: 2507.10977 by the authors.

**Figure 1.** Comparison between (a) existing architecture and (b) proposed architecture in extracting features. The heatmaps are generated using the last layer [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗

**Figure 2.** Wavelet backbone architecture is divided into 2 stages, feature extraction and feature refinement, to obtain focused, rich semantic features. Note that [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗

**Figure 3.** Proposed ray mechanism. 12 ray origins are utilized to produce a [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗

Figures from the paper (4 more)

**Figure 4.** Figure 4: Grad-CAM activation maps on (a) Wavelet and (b) Wavelet+3 Rays. [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]

**Figure 6.** Figure 6: Visualization of Grad-CAM activation maps on different test images [PITH_FULL_IMAGE:figures/full_fig_p006_6.png]

**Figure 7.** Figure 7: Eigen-CAM [45] activation maps on (a) Swin, (b) wavelet, and (c) [PITH_FULL_IMAGE:figures/full_fig_p007_7.png]

**Figure 8.** Figure 8: Visualization of Eigen-CAM activation maps on challenging test [PITH_FULL_IMAGE:figures/full_fig_p007_8.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

45 extracted references · 21 canonical work pages

[1]

End-to-end object detection with transformers,

N. Carion, F. Massa, G. Synnaeve, N. Usunier, A. Kirillov, and S. Zagoruyko, “End-to-end object detection with transformers,” in European conference on computer vision . Springer, 2020, pp. 213– 229

2020
[2]

You only look once: Unified, real-time object detection,

J. Redmon, S. Divvala, R. Girshick, and A. Farhadi, “You only look once: Unified, real-time object detection,” in Proceedings of the IEEE conference on computer vision and pattern recognition , 2016, pp. 779– 788

2016
[3]

Real-time end-to-end action detection with two-stream networks,

A. Ali and G. W. Taylor, “Real-time end-to-end action detection with two-stream networks,” in 2018 15th Conference on Computer and Robot Vision (CRV). IEEE, 2018, pp. 31–38

work page 2018
[4]

With a Little Help from my Temporal Context: Multimodal Egocentric Action Recognition

E. Kazakos, J. Huh, A. Nagrani, A. Zisserman, and D. Damen, “With a little help from my temporal context: Multimodal egocentric action recognition,” arXiv preprint arXiv:2111.01024 , 2021

work page Pith review arXiv 2021
[5]

Hierarchical reasoning network for human-object interaction detection,

Y . Gao, Z. Kuang, G. Li, W. Zhang, and L. Lin, “Hierarchical reasoning network for human-object interaction detection,” IEEE Transactions on Image Processing, vol. 30, pp. 8306–8317, 2021

2021
[6]

Hier r-cnn: Instance- level human parts detection and a new benchmark,

L. Yang, Q. Song, Z. Wang, M. Hu, and C. Liu, “Hier r-cnn: Instance- level human parts detection and a new benchmark,” IEEE Transactions on Image Processing , vol. 30, pp. 39–54, 2020

work page 2020
[7]

Reformu- lating hoi detection as adaptive set prediction,

M. Chen, Y . Liao, S. Liu, Z. Chen, F. Wang, and C. Qian, “Reformu- lating hoi detection as adaptive set prediction,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , 2021, pp. 9004–9013

2021
[8]

Image captioning with semantic attention,

Q. You, H. Jin, Z. Wang, C. Fang, and J. Luo, “Image captioning with semantic attention,” in Proceedings of the IEEE conference on computer vision and pattern recognition , 2016, pp. 4651–4659

2016

Show all 45 references

[9]

Vqa: Visual question answering,

S. Antol, A. Agrawal, J. Lu, M. Mitchell, D. Batra, C. L. Zitnick, and D. Parikh, “Vqa: Visual question answering,” inProceedings of the IEEE international conference on computer vision , 2015, pp. 2425–2433

2015
[10]

Human- object interaction: Application to abandoned luggage detection in video surveillance scenarios,

M. Dogariu, L.-D. Stefan, M. G. Constantin, and B. Ionescu, “Human- object interaction: Application to abandoned luggage detection in video surveillance scenarios,” in 2020 13th International Conference on Com- munications (COMM). IEEE, 2020, pp. 157–160

2020
[11]

A survey of autonomous driving: Common practices and emerging technologies,

E. Yurtsever, J. Lambert, A. Carballo, and K. Takeda, “A survey of autonomous driving: Common practices and emerging technologies,” IEEE access, vol. 8, pp. 58 443–58 469, 2020

2020
[12]

Deformable detr: Deformable transformers for end-to-end object detection,

X. Zhu, W. Su, L. Lu, B. Li, X. Wang, and J. Dai, “Deformable detr: Deformable transformers for end-to-end object detection,”arXiv preprint arXiv:2010.04159, 2020

2010 arXiv
[13]

Deep residual learning for image recognition,

K. He, X. Zhang, S. Ren, and J. Sun, “Deep residual learning for image recognition,” in Proceedings of the IEEE conference on computer vision and pattern recognition , 2016, pp. 770–778

2016
[14]

Swin transformer: Hierarchical vision transformer using shifted windows,

Z. Liu, Y . Lin, Y . Cao, H. Hu, Y . Wei, Z. Zhang, S. Lin, and B. Guo, “Swin transformer: Hierarchical vision transformer using shifted windows,” in Proceedings of the IEEE/CVF international conference on computer vision, 2021, pp. 10 012–10 022

2021
[15]

Qahoi: Query-based anchors for human-object interaction detection,

J. Chen and K. Yanai, “Qahoi: Query-based anchors for human-object interaction detection,” in 2023 18th International Conference on Ma- chine Vision and Applications (MVA) . IEEE, 2023, pp. 1–5

2023
[16]

Ernet: An efficient and reliable human-object interaction detection network,

J. Lim, V . M. Baskaran, J. M.-Y . Lim, K. Wong, J. See, and M. Tistarelli, “Ernet: An efficient and reliable human-object interaction detection network,” IEEE Transactions on Image Processing , vol. 32, pp. 964– 979, 2023

2023
[17]

Learning transferable human-object interaction detector with natural language supervision,

S. Wang, Y . Duan, H. Ding, Y .-P. Tan, K.-H. Yap, and J. Yuan, “Learning transferable human-object interaction detector with natural language supervision,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , 2022, pp. 939–948

2022
[18]

Rlip: Relational language-image pre-training for human-object inter- action detection,

H. Yuan, J. Jiang, S. Albanie, T. Feng, Z. Huang, D. Ni, and M. Tang, “Rlip: Relational language-image pre-training for human-object inter- action detection,” Advances in Neural Information Processing Systems , vol. 35, pp. 37 416–37 431, 2022

2022
[19]

Spatially conditioned graphs for detecting human-object interactions,

F. Z. Zhang, D. Campbell, and S. Gould, “Spatially conditioned graphs for detecting human-object interactions,” in Proceedings of the IEEE/CVF International Conference on Computer Vision , 2021, pp. 13 319–13 327

2021
[20]

Pose graph parsing network for human-object interaction detection,

Z. Su, Y . Wang, Q. Xie, and R. Yu, “Pose graph parsing network for human-object interaction detection,” Neurocomputing, vol. 476, pp. 53– 62, 2022

2022
[21]

A convnet for the 2020s,

Z. Liu, H. Mao, C.-Y . Wu, C. Feichtenhofer, T. Darrell, and S. Xie, “A convnet for the 2020s,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , 2022, pp. 11 976–11 986

2022
[22]

Discovering and explaining the representation bottleneck of dnns,

H. Deng, Q. Ren, H. Zhang, and Q. Zhang, “Discovering and explaining the representation bottleneck of dnns,” arXiv preprint arXiv:2111.06236, 2021

2021 arXiv
[23]

Fgahoi: Fine-grained anchors for human-object interaction detection,

S. Ma, Y . Wang, S. Wang, and Y . Wei, “Fgahoi: Fine-grained anchors for human-object interaction detection,” IEEE Transactions on Pattern Analysis and Machine Intelligence , 2023

2023
[24]

Faster r-cnn: Towards real-time object detection with region proposal networks,

S. Ren, K. He, R. Girshick, and J. Sun, “Faster r-cnn: Towards real-time object detection with region proposal networks,” Advances in neural information processing systems , vol. 28, 2015

2015
[25]

Learning human-object interactions by graph parsing neural networks,

S. Qi, W. Wang, B. Jia, J. Shen, and S.-C. Zhu, “Learning human-object interactions by graph parsing neural networks,” in Proceedings of the European conference on computer vision (ECCV) , 2018, pp. 401–417

2018
[26]

Visual compositional learning for human-object interaction detection,

Z. Hou, X. Peng, Y . Qiao, and D. Tao, “Visual compositional learning for human-object interaction detection,” in Computer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, Proceedings, Part XV 16 . Springer, 2020, pp. 584–600

2020
[27]

Uniondet: Union-level detec- tor towards real-time human-object interaction detection,

B. Kim, T. Choi, J. Kang, and H. J. Kim, “Uniondet: Union-level detec- tor towards real-time human-object interaction detection,” in Computer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, Proceedings, Part XV 16 . Springer, 2020, pp. 498–514

2020
[28]

Learning transferable visual models from natural language supervision,

A. Radford, J. W. Kim, C. Hallacy, A. Ramesh, G. Goh, S. Agarwal, G. Sastry, A. Askell, P. Mishkin, J. Clark et al., “Learning transferable visual models from natural language supervision,” in International conference on machine learning . PMLR, 2021, pp. 8748–8763

2021
[29]

Dual- branch knowledge enhancement network with vision-language model for human-object interaction detection,

G. Zhou, D. Kong, J. Li, D. Chen, Z. Bai, and B. Yin, “Dual- branch knowledge enhancement network with vision-language model for human-object interaction detection,” in 2024 International Joint Conference on Neural Networks (IJCNN) . IEEE, 2024, pp. 1–8

2024
[30]

Deep adaptive wavelet network,

M. X. B. Rodriguez, A. Gruson, L. Polania, S. Fujieda, F. Prieto, K. Takayama, and T. Hachisuka, “Deep adaptive wavelet network,” in Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, 2020, pp. 3111–3119

2020
[31]

Wavelet convolutional neural networks,

S. Fujieda, K. Takayama, and T. Hachisuka, “Wavelet convolutional neural networks,” 2018

2018
[32]

Learning-based image compression using convolutional autoencoder and wavelet decomposition,

P. Akyazi and T. Ebrahimi, “Learning-based image compression using convolutional autoencoder and wavelet decomposition,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recogni- tion (CVPR) Workshops, June 2019

2019
[33]

Wavelet-srnet: A wavelet-based cnn for multi-scale face super resolution,

H. Huang, R. He, Z. Sun, and T. Tan, “Wavelet-srnet: A wavelet-based cnn for multi-scale face super resolution,” in Proceedings of the IEEE International Conference on Computer Vision (ICCV) , Oct 2017

2017
[34]

Winnet: Wavelet-inspired invertible network for image denoising,

J.-J. Huang and P. L. Dragotti, “Winnet: Wavelet-inspired invertible network for image denoising,” IEEE Transactions on Image Processing, vol. 31, pp. 4377–4392, 2022

2022
[35]

An explainable spatial-frequency multi-scale transformer for remote sensing scene classification,

Y . Yang, L. Jiao, F. Liu, X. Liu, L. Li, P. Chen, and S. Yang, “An explainable spatial-frequency multi-scale transformer for remote sensing scene classification,” IEEE Transactions on Geoscience and Remote Sensing, 2023

2023
[36]

Light field networks: Neural scene representations with single- evaluation rendering,

V . Sitzmann, S. Rezchikov, B. Freeman, J. Tenenbaum, and F. Du- rand, “Light field networks: Neural scene representations with single- evaluation rendering,” Advances in Neural Information Processing Sys- tems, vol. 34, pp. 19 313–19 325, 2021

2021
[37]

Point- ersect: Neural rendering with cloud-ray intersection,

J.-H. R. Chang, W.-Y . Chen, A. Ranjan, K. M. Yi, and O. Tuzel, “Point- ersect: Neural rendering with cloud-ray intersection,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023, pp. 8359–8369

2023
[38]

Generalization and network design strategies,

Y . LeCun et al. , “Generalization and network design strategies,” Con- nectionism in perspective , vol. 19, no. 143-155, p. 18, 1989

1989
[39]

Conv2former: A simple transformer-style convnet for visual recognition,

Q. Hou, C.-Z. Lu, M.-M. Cheng, and J. Feng, “Conv2former: A simple transformer-style convnet for visual recognition,” arXiv preprint arXiv:2211.11943, 2022

2022 arXiv
[40]

Attention is all you need,

A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, Ł. Kaiser, and I. Polosukhin, “Attention is all you need,” Advances in neural information processing systems , vol. 30, 2017

2017
[41]

Global filter networks for image classification,

Y . Rao, W. Zhao, Z. Zhu, J. Lu, and J. Zhou, “Global filter networks for image classification,” Advances in neural information processing systems, vol. 34, pp. 980–993, 2021

2021
[42]

Training data-efficient image transformers & distillation through attention,

H. Touvron, M. Cord, M. Douze, F. Massa, A. Sablayrolles, and H. J ´egou, “Training data-efficient image transformers & distillation through attention,” in International conference on machine learning . PMLR, 2021, pp. 10 347–10 357

2021
[43]

Grad-cam: Visual explanations from deep networks via gradient-based localization,

R. R. Selvaraju, M. Cogswell, A. Das, R. Vedantam, D. Parikh, and D. Batra, “Grad-cam: Visual explanations from deep networks via gradient-based localization,” in Proceedings of the IEEE international conference on computer vision , 2017, pp. 618–626

2017
[44]

Deep feature factorization for concept discovery,

E. Collins, R. Achanta, and S. Susstrunk, “Deep feature factorization for concept discovery,” in Proceedings of the European Conference on Computer Vision (ECCV) , 2018, pp. 336–352

2018
[45]

Eigen-cam: Class activation map using principal components,

M. B. Muhammad and M. Yeasin, “Eigen-cam: Class activation map using principal components,” in 2020 international joint conference on neural networks (IJCNN) . IEEE, 2020, pp. 1–7

2020

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.

[1] [1]

End-to-end object detection with transformers,

N. Carion, F. Massa, G. Synnaeve, N. Usunier, A. Kirillov, and S. Zagoruyko, “End-to-end object detection with transformers,” in European conference on computer vision . Springer, 2020, pp. 213– 229

2020

[2] [2]

You only look once: Unified, real-time object detection,

J. Redmon, S. Divvala, R. Girshick, and A. Farhadi, “You only look once: Unified, real-time object detection,” in Proceedings of the IEEE conference on computer vision and pattern recognition , 2016, pp. 779– 788

2016

[3] [3]

Real-time end-to-end action detection with two-stream networks,

A. Ali and G. W. Taylor, “Real-time end-to-end action detection with two-stream networks,” in 2018 15th Conference on Computer and Robot Vision (CRV). IEEE, 2018, pp. 31–38

work page 2018

[4] [4]

With a Little Help from my Temporal Context: Multimodal Egocentric Action Recognition

E. Kazakos, J. Huh, A. Nagrani, A. Zisserman, and D. Damen, “With a little help from my temporal context: Multimodal egocentric action recognition,” arXiv preprint arXiv:2111.01024 , 2021

work page Pith review arXiv 2021

[5] [5]

Hierarchical reasoning network for human-object interaction detection,

Y . Gao, Z. Kuang, G. Li, W. Zhang, and L. Lin, “Hierarchical reasoning network for human-object interaction detection,” IEEE Transactions on Image Processing, vol. 30, pp. 8306–8317, 2021

2021

[6] [6]

Hier r-cnn: Instance- level human parts detection and a new benchmark,

L. Yang, Q. Song, Z. Wang, M. Hu, and C. Liu, “Hier r-cnn: Instance- level human parts detection and a new benchmark,” IEEE Transactions on Image Processing , vol. 30, pp. 39–54, 2020

work page 2020

[7] [7]

Reformu- lating hoi detection as adaptive set prediction,

M. Chen, Y . Liao, S. Liu, Z. Chen, F. Wang, and C. Qian, “Reformu- lating hoi detection as adaptive set prediction,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , 2021, pp. 9004–9013

2021

[8] [8]

Image captioning with semantic attention,

Q. You, H. Jin, Z. Wang, C. Fang, and J. Luo, “Image captioning with semantic attention,” in Proceedings of the IEEE conference on computer vision and pattern recognition , 2016, pp. 4651–4659

2016