Pith. sign in

REVIEW 4 major objections 4 minor 31 references

Great Ape Detection in Challenging Jungle Camera Trap Footage via Attention-Based Spatial and Temporal Feature Blending

T0 review · 4 major / 4 minor · reviewed 2026-08-14 · deepseek-v4-flash

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

Pith's one-line read Attention-based blending across space and time lifts great-ape detection in jungle camera traps to 91.17% mean average precision.

desk verdict Useful great ape camera trap dataset and a plausible attention-based video detector, but the headline mAP is selected on the same test set it's reported on, so the generalization claim needs a proper validation split. read the letter →

arxiv 1908.11240 v1 pith:AGOZXB7U submitted 2019-08-29 cs.CV

Xinyu Yang , Majid Mirmehdi , Tilo Burghardt This is my paper · ORCID
classification cs.CV
keywords videoobjectdetectiongreatapecameratrapfootageself-attentionspatialfeatureblendingtemporalpyramidnetworkocclusionrobustness
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's aim is to show that video, not just single frames, is the right unit for detecting great apes in difficult jungle camera-trap footage. It introduces two self-attention modules that blend spatial and temporal features inside a standard feature-pyramid detector, and it reports that this addition lifts mean average precision from about 81% to 91% on a 500-clip, 180K-frame manually annotated great-ape dataset. If true, the result matters because occlusion and poor lighting are exactly the conditions that defeat frame-by-frame detection in real camera-trap surveys, and explicit per-frame locations would directly assist human review.

What carries the argument

The load-bearing object is the attention-based blending module, used in two variants. The SCM computes a softmax attention map over spatial positions and multiplies it back into the feature map; the TCM computes a softmax over a temporal window of $T$ frames and blends the attention-weighted features with learned weights, then adds the result back as a residual. Inserted after the final residual addition of the backbone and feeding into the feature pyramid network, these modules let the detector reconstruct an animal's location from dispersed spatial and temporal cues.

What would settle it

Compute the headline mAP on a test split that is frozen before any ablation, with all design choices fixed; if the 90-91% figure is not reproduced, or if independent re-annotation of a sample of the 180K boxes shows low inter-annotator agreement, the claimed readiness to assist inspection would not be supported.

Watch

Extended reading notes

Core claim

The central claim, stated the way a sympathetic author would state it, is that a feature-pyramid object detector for video can be made substantially more robust to occlusion and harsh lighting by inserting two trained attention modules: a Spatial Context Module (SCM) that blends informative spatial locations within a frame, and a Temporal Context Module (TCM) that blends features across a short window of neighbouring frames. The paper reports that this SCM+TCM extension improves detection of great apes in jungle camera-trap footage from 80.79% to 90.81% mAP on a ResNet50 RetinaNet baseline, and to 91.17% mAP with a Cascade head, and that it also performs well on a general video detection benchmark. It concludes that the system is ready to assist, not replace, human inspection.

Load-bearing premise

The reported accuracy rests on the assumption that tuning design choices on the same 100-video test set that later produces the headline score does not inflate that score, together with the unquantified reliability of the manual bounding-box labels.

Editorial extensions

If this is right

  • The SCM+TCM detector localises great apes in frames where single-frame detectors fail, specifically under partial occlusion and low light.
  • Adding the modules to a ResNet50 RetinaNet backbone raises mAP from 80.79% to 90.81% on the Pan Africa test set, and the best Cascade-based version reaches 91.17%.
  • Widening the temporal window at test time improves accuracy even when training used shorter snippets, by about two to four mAP points.
  • The same architecture reaches 63.85% mAP on a general video-object-detection corpus, showing the extension is not limited to apes.
  • With published code, weights, and 180K-frame annotations, the framework is positioned as practical assistance for human camera-trap review rather than a replacement.

Reading between the lines

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

  • If the gain is real beyond this test set, the same SCM+TCM add-on could improve detection for other partially occluded species in camera-trap surveys, since the modules are not ape-specific.
  • The ablation's finding that longer test-time temporal windows improve accuracy suggests a zero-retraining deployment trick: feed the trained model longer snippets at inference to gain several mAP points.
  • Because the paper publishes weights and annotations, an independent test can directly measure how sensitive the result is to annotation noise by re-annotating a sample and comparing the labels.
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.

Referee Report

4 major / 4 minor

Summary. The paper proposes an extension to feature-pyramid object detectors for video: a Spatial Context Module (SCM) and a Temporal Context Module (TCM), both driven by self-attention, which blend spatial and temporal features before detection heads. The method is evaluated on a newly annotated Pan Africa camera-trap dataset of 500 clips (180K frames) of great apes, using RetinaNet and Cascade-RCNN backbones, and additionally validated on ILSVRC 2015 VID. The main reported results show mAP improvements from 80.79/85.25/88.31 (frame baselines) to 90.81/90.21/91.17 with SCM+TCM on Pan Africa, with ablation studies on insertion point, temporal window, and embedding strategy. The authors conclude that the framework is ready to assist human camera-trap inspection efforts and release code, network weights, and ground-truth annotations.

Significance. If the reported performance generalizes, the contribution is practically valuable: attention-based spatial and temporal feature blending is a simple, modular add-on to standard FPN detectors and appears to help under occlusion and poor lighting, which is exactly the regime where camera-trap animal detection struggles. The paper has clear strengths: it releases code, weights, and 180K-frame annotations; it re-implements and compares against reasonably strong frame-based baselines; it provides qualitative analyses; and it validates on a second large dataset (ILSVRC 2015 VID). These are real assets. However, as submitted, the central quantitative claim is not yet an unbiased generalization estimate: the same 100 test clips are used both to select design choices and to report the headline mAP, no variance or significance testing is reported, no comparison to prior video-detection methods or to non-attention temporal aggregation is given, and annotation quality is unquantified. The evaluation protocol, not the architecture itself, is the main obstacle to accepting the paper's readiness claim.

major comments (4)
  1. [Section 4.3, Tables 2(a)-(c), Table 1] The ablation study is performed on the same Pan Africa test set of 100 clips that is later used to report the headline mAP values (90.81--91.17% in Table 1). Insertion point, T_train=7/T_test=21, and the Main & Refs embedding strategy are all selected from the results in Tables 2(a)--(c), so the final numbers are optimistically biased. An independent validation split, a nested evaluation, or at minimum a site-aware split is required before the reported mAP can be read as an unbiased estimate of performance on unseen clips. This is load-bearing because the conclusion that the system is 'ready to assist' rests on these numbers.
  2. [Sections 4.1 and 5] The paper compares only against re-implemented single-frame RetinaNet and Cascade-RCNN baselines; Section 5 explicitly states that tests against other current video detection frameworks are outstanding. Without comparisons to existing video-object-detection methods (e.g., FGFA, DFF, ST-Lattice, Seq-NMS, or T-CNN) the claim that the proposed approach is a strong video detector is unsupported. Further, no ablation against a non-attention temporal feature aggregation (such as simple averaging or flow-based aggregation) is provided, so the attributed benefit of 'attention-based' blending over generic temporal context is unmeasured.
  3. [Tables 1-3] No variance, confidence intervals, or significance tests are reported; each configuration appears to be a single run. Differences as small as 0.16 mAP (e.g., Table 2(c): Positional 88.61 vs Main & Refs 90.81) are discussed as improvements, and the abstract uses the word 'significantly' without a statistical test. Multi-seed runs, per-clip bootstrap intervals, or paired tests are needed to support the strength of the claims.
  4. [Contributions (Section 1) and Section 4.1] The paper introduces 180K manually annotated bounding boxes as a contribution, but reports no inter-annotator agreement, label-quality audit, or annotation error analysis. Given that label noise directly affects mAP and the dataset is released for community use, the absence of any reliability measure weakens both the benchmark and the interpretation of the reported differences between methods.
minor comments (4)
  1. [Figure 4 caption] The word 'indicte' appears in the caption; it should be 'indicate'.
  2. [Section 3, Temporal Module] The phrase 'principles layed out' should be 'principles laid out'.
  3. [Section 3.2 vs Table 1] The text says training snippets of T=3 to T=8 are used, while Table 1 reports T_train values of 3, 5, and 7; please clarify the exact training configuration for each row.
  4. [Section 3, Equations (3)-(5)] The notation for w4_t, w5_m, and w6_m is introduced only at a high level; please specify the tensor shapes and where each transformation is applied so that the modules can be reproduced exactly.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the paper reports an empirical detection evaluation with explicit network equations; the test-set model selection concern is a generalization issue, not a circular derivation.

full rationale

This is an empirical computer-vision evaluation rather than a formal derivation, and no load-bearing step reduces to its own inputs by construction. The SCM and TCM components are specified by explicit equations (2)-(5) in terms of learnable self-attention weights and are trained end-to-end; the reported mAP figures are measurements on a held-out portion of the Pan Africa data (500 clips, with 400 used for training and 100 for testing) plus independent validation on the full ILSVRC 2015 VID corpus. Citations to GCNet, RetinaNet, and Cascade R-CNN are implementation references, not unverified premises that determine the outcome. The closest concern is that the Section 4.3 ablation choices (insertion point, T_train/T_test, and embedding strategy) are selected using the same 100-clip Pan Africa test set that yields the Table 1 headline mAP; this is a statistical generalization or overfitting concern, not argumentative circularity, because the headline number is not a fitted parameter renamed as a prediction and no equation is equivalent to its input by definition. Section 5 also explicitly concedes that comparisons to other video detection frameworks are outstanding, which is an acknowledged empirical gap rather than a circular step. Therefore no circularity is established, and the score is 0.

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

The paper's central claim is empirical and rests on learned network parameters plus several unstated evaluation assumptions. We list the hand-set design choices that the final mAP depends on, the ground-truth quality assumption, the test-set reuse issue, and the transfer-learning assumption. There is no parameter-free derivation; all numbers reported are measured on data.

free parameters (5)
  • Snippet length during training (T_train) = 7 for final Pan Africa model; 3, 5, 8 tested
    Temporal context size is a hand-set hyperparameter chosen by ablation; larger T improved mAP but was limited by GPU memory. It directly controls the temporal attention range in Equations (3) to (5).
  • Test-time temporal window (T_test) = 21 for reported best results; 5 and 8 compared
    Widening the test window from 5 to 21 gained 1.96% mAP on Pan Africa, so the final claim uses the best window.
  • SCM/TCM insertion point = After the residual addition of the last ResNet block
    Three insertion locations were tested on the test set; after-add gave the highest mAP and is used in all final numbers.
  • TCM embedding strategy = Main and Refs (only main frame backpropagated)
    Ablation Table 2(c) showed 90.81% versus 88.61% for the positional variant; the better variant is used in the headline.
  • All network weights of SCM, TCM, FPN, and detection heads = Trained end-to-end on the Pan Africa training split after ImageNet VID pretraining
    The attention modules have learnable parameters w1 through w6, and all detection weights are optimized on data; these are the fitted quantities the method depends on.
assumptions (4)
  • domain assumption The 180K manual bounding boxes in the Pan Africa dataset are accurate enough to serve as ground truth.
    Labels were produced by the authors; no inter-annotator agreement or independent spot-check is reported (Sections 1 and 4.1).
  • ad hoc to paper The 100-clip Pan Africa test split can be used both to select architecture variants and to report final performance without meaningful overfitting.
    Ablation choices in Section 4.3 are evaluated on the same test set that produces Table 1, an optimistic assumption for the reported mAP.
  • domain assumption Pretraining on ImageNet VID transfers to jungle camera trap footage.
    Two-stage training in Section 3.2 relies on this transfer; Figure 5 shows validation mAP curves but no domain-shift analysis.
  • domain assumption Features in neighboring frames within a short snippet contain recoverable cues for partially occluded apes.
    This is the core inductive bias of the TCM behind Equation (5); it is supported by attention map visualizations but not by an independent failure analysis.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Great Ape Detection in Challenging Jungle Camera Trap Footage via Attention-Based Spatial and Temporal Feature Blending." pith.science (2026). https://pith.science/paper/AGOZXB7U

@misc{pith2026190811240,
  author       = {Pith},
  title        = {Pith review of: Great Ape Detection in Challenging Jungle Camera Trap Footage via Attention-Based Spatial and Temporal Feature Blending},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/AGOZXB7U}},
  note         = {Machine review of arXiv:1908.11240}
}
read the original abstract

We propose the first multi-frame video object detection framework trained to detect great apes. It is applicable to challenging camera trap footage in complex jungle environments and extends a traditional feature pyramid architecture by adding self-attention driven feature blending in both the spatial as well as the temporal domain. We demonstrate that this extension can detect distinctive species appearance and motion signatures despite significant partial occlusion. We evaluate the framework using 500 camera trap videos of great apes from the Pan African Programme containing 180K frames, which we manually annotated with accurate per-frame animal bounding boxes. These clips contain significant partial occlusions, challenging lighting, dynamic backgrounds, and natural camouflage effects. We show that our approach performs highly robustly and significantly outperforms frame-based detectors. We also perform detailed ablation studies and validation on the full ILSVRC 2015 VID data corpus to demonstrate wider applicability at adequate performance levels. We conclude that the framework is ready to assist human camera trap inspection efforts. We publish code, weights, and ground truth annotations with this paper.

Figures

Figures reproduced from arXiv: 1908.11240 by the authors.

Figure 1
Figure 1. Architectural Overview. Our framework processes video snippets (top left) and robustly detects per-frame great ape locations (bottom right). The architecture extends a traditional de￾tection backbone (blue) by two self-attention components, which select and ‘blend’ most relevant temporal (TCM, red) and spatial (SCM, green) information via SoftMax folding. Blended feature maps are fed back into the feature pyramid ne… view at source ↗
Figure 2
Figure 2. exemplifies such scenarios on the Pan Africa cam￾era trap data used in this work. This dataset contains 500× 15s video clips with the resolution of 720×404 and was col￾lected by the Pan African Programme ‘The Cultured Chim￾panzee’ (see Acknowledgements). It was subsequently la￾belled for this paper with accurate bounding box ground truth for all animals in each frame. Contributions. (1) FPN-integrated blended detect… view at source ↗
Figure 3
Figure 3. Principled Workflow of the Blending Framework. (1) Input features are grouped by the target dimension; (2) Each location of the feature map is embedded by its position; (3) A cor￾relation map is computed and reduced. This map is then applied back to the embedded feature via matrix multiplication (marked as N) implementing the self-attention principle. (4) The result￾ing feature is finally transformed, before (4) bei… view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: Visualisation of Temporal Attention Maps for Two Example Scenes. First and third rows depict attention maps when projected back into the image domain. Second and fourth row show the associated frames with green bounding boxes showing predictions by the network with T =…
Figure 5
Figure 5. Figure 5: Example Training and Validation Performance. mAP plots for both training (blue) and validation (green) of the ResNet50 RetinaNet SCM+TCM setup along the 28 epochs of op￾timisation split into two distinct phases, i.e. pre-training on Ima￾geNet VID, and fine-tuning on th…
Figure 6
Figure 6. Figure 6: Examples of Positive Attention Component Impact. Depicted are ground truth labels (red) and detections (green) for a ResNet50 FPN with (rows 2, 4, and 6) and without (rows 1, 3, and 5) SCM+TCM components. Note the clear detection im￾provements for partially occluded gr…
Figure 7
Figure 7. Figure 7: Success and Failure Cases. Depicted are ground truth labels (red) and detections (green) in frames of various ex￾ample clips using the best performing model ResX101 Cascade with SCM+TCM components [PITH_FULL_IMAGE:figures/full_fig_p006_7.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

31 extracted references · 28 canonical work pages

  1. [1]

    Henriques, Andrea Vedaldi, and Philip H.S

    Luca Bertinetto, Jack Valmadre, Joo F. Henriques, Andrea Vedaldi, and Philip H.S. Torr. Fully-convolutional siamese networks for object tracking. In ECCV, pages 850–865, 2016

    work page 2016

  2. [2]

    Towards Automated Vi- sual Monitoring of Individual Gorillas in the Wild

    Clemens-Alexander Brust, Tilo Burghardt, Milou Groe- nenberg, Christoph K ¨ading, Hjalmar S K ¨uhl, Marie L Manguette, and Joachim Denzler. Towards Automated Vi- sual Monitoring of Individual Gorillas in the Wild. In ICCVW, pages 2820–2830, 2017

    work page 2017

  3. [3]

    Cascade R-CNN: Delv- ing into High Quality Object Detection

    Zhaowei Cai and Nuno Vasconcelos. Cascade R-CNN: Delv- ing into High Quality Object Detection. In CVPR, pages 6154–6162, 2018

    work page 2018

  4. [4]

    GCNet: Non-local Networks Meet Squeeze-Excitation Networks and Beyond

    Yue Cao, Jiarui Xu, Stephen Lin, Fangyun Wei, and Han Hu. GCNet: Non-local Networks Meet Squeeze-Excitation Networks and Beyond. arXiv:1904.11492 [cs], 2019

    arXiv 1904

  5. [5]

    Quo Vadis, action recognition? A new model and the kinetics dataset

    Joo Carreira and Andrew Zisserman. Quo Vadis, action recognition? A new model and the kinetics dataset. In CVPR, pages 4724–4733, 2017

    work page 2017

  6. [6]

    Optimizing Video Object Detection via a Scale-Time Lattice

    Kai Chen, Jiaqi Wang, Shuo Yang, Xingcheng Zhang, Yuan- jun Xiong, Chen Change Loy, and Dahua Lin. Optimizing Video Object Detection via a Scale-Time Lattice. In CVPR, pages 7814–7823, 2018

    work page 2018

  7. [7]

    Auto- mated face detection for occurrence and occupancy esti- mation in chimpanzees

    Anne-Sophie Crunchant, Monika Egerer, Alexander Loos, Tilo Burghardt, Klaus Zuberb ¨uhler, Katherine Corogenes, Vera Leinert, Lars Kulik, and Hjalmar S K ¨uhl. Auto- mated face detection for occurrence and occupancy esti- mation in chimpanzees. American journal of primatology, 79(3):e22627, 2017

    work page 2017

  8. [8]

    Competition: Pri-matrix Factorization [online]

    DrivenData. Competition: Pri-matrix Factorization [online]. https://www.drivendata.org/competitions/ 49/deep-learning-camera-trap-animals/ . Accessed: 2019-07-30

    work page 2019

Show all 31 references
  1. [9]

    Detect to Track and Track to Detect

    Christoph Feichtenhofer, Axel Pinz, and Andrew Zisserman. Detect to Track and Track to Detect. In ICCV, pages 3038– 3046, 2017

    2017

  2. [10]

    Wei Han, Pooya Khorrami, Tom Le Paine, Prajit Ramachan- dran, Mohammad Babaeizadeh, Honghui Shi, Jianan Li, Shuicheng Yan, and Thomas S. Huang. Seq-NMS for Video Object Detection. Technical report, 2016

    2016

  3. [11]

    Deep residual learning for image recognition

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. In CVPR, pages 770–778, 2016

    2016

  4. [12]

    ImageNet Challenge 2015: Object Detec- tion from Video [online]

    ImageNet. ImageNet Challenge 2015: Object Detec- tion from Video [online]. http://image-net.org/ challenges/LSVRC/2015/index#vid. Accessed: 2019-07-30

    2015

  5. [13]

    T-CNN: Tubelets with Convo- lutional Neural Networks for Object Detection from Videos

    Kai Kang, Hongsheng Li, Junjie Yan, Xingyu Zeng, Bin Yang, Tong Xiao, Cong Zhang, Ruohui Wang, Xiaogang Wang, and Wanli Ouyang. T-CNN: Tubelets with Convo- lutional Neural Networks for Object Detection from Videos. IEEE T-CSVT, 28(10):2896–2907, 2018

    2018

  6. [14]

    Object detection from video tubelets with convolu- tional neural networks

    Kai Kang, Wanli Ouyang, Hongsheng Li, and Xiaogang Wang. Object detection from video tubelets with convolu- tional neural networks. In CVPR, pages 817–825, 2016

    2016

  7. [15]

    Large-scale video classification with convolutional neural networks

    Andrej Karpathy, George Toderici, Sanketh Shetty, Thomas Leung, Rahul Sukthankar, and Fei Fei Li. Large-scale video classification with convolutional neural networks. In CVPR, pages 1725–1732, 2014

    2014

  8. [16]

    Kuehl and Tilo Burghardt

    Hjalmar S. Kuehl and Tilo Burghardt. Animal biometrics: quantifying and detecting phenotypic appearance. TREE, 28(7):432 – 441, 2013

    2013

  9. [17]

    SiamRPN++: Evolution of Siamese Vi- sual Tracking with Very Deep Networks.arXiv:1812.11703, 2018

    Bo Li, Wei Wu, Qiang Wang, Fangyi Zhang, Junliang Xing, and Junjie Yan. SiamRPN++: Evolution of Siamese Vi- sual Tracking with Very Deep Networks.arXiv:1812.11703, 2018

    2018 arXiv

  10. [18]

    Feature pyramid networks for object detection

    Tsung Yi Lin, Piotr Doll ´ar, Ross Girshick, Kaiming He, Bharath Hariharan, and Serge Belongie. Feature pyramid networks for object detection. In CVPR, pages 936–944, 2017

    2017

  11. [19]

    Focal Loss for Dense Object Detection

    Tsung Yi Lin, Priya Goyal, Ross Girshick, Kaiming He, and Piotr Dollar. Focal Loss for Dense Object Detection. In ICCV, pages 2999–3007, 2017

    2017

  12. [20]

    Wei Liu, Dragomir Anguelov, Dumitru Erhan, Christian Szegedy, Scott Reed, Cheng Yang Fu, and Alexander C. Berg. SSD: Single shot multibox detector. In ECCV, vol- ume 9905 LNCS, pages 21–37, 2016

    2016

  13. [21]

    An automated chim- panzee identification system using face detection and recog- nition

    Alexander Loos and Andreas Ernst. An automated chim- panzee identification system using face detection and recog- nition. EURASIP Journal on Image and Video Processing, 2013(1):49, 2013

    2013

  14. [22]

    You only look once: Unified, real-time object de- tection

    Joseph Redmon, Santosh Divvala, Ross Girshick, and Ali Farhadi. You only look once: Unified, real-time object de- tection. In CVPR, pages 779–788, 2016

    2016

  15. [23]

    YOLO9000: Better, faster, stronger

    Joseph Redmon and Ali Farhadi. YOLO9000: Better, faster, stronger. In CVPR, pages 6517–6525, 2017

    2017

  16. [24]

    Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks

    Shaoqing Ren, Kaiming He, Ross Girshick, and Jian Sun. Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks. PAMI, 39(6):1137–1149, 2017

    2017

  17. [25]

    Two-Stream Con- volutional Networks for Action Recognition in Videos

    Karen Simonyan and Andrew Zisserman. Two-Stream Con- volutional Networks for Action Recognition in Videos. In NIPS, pages 568–576, 2014

    2014

  18. [26]

    Learning spatiotemporal features with 3D convolutional networks

    Du Tran, Lubomir Bourdev, Rob Fergus, Lorenzo Torresani, and Manohar Paluri. Learning spatiotemporal features with 3D convolutional networks. In ICCV, pages 4489–4497, 2015

    2015

  19. [27]

    Temporal segment networks: Towards good practices for deep action recogni- tion

    Limin Wang, Yuanjun Xiong, Zhe Wang, Yu Qiao, Dahua Lin, Xiaoou Tang, and Luc van Gool. Temporal segment networks: Towards good practices for deep action recogni- tion. In ECCV, pages 20–36, 2016

    2016

  20. [28]

    Non-local Neural Networks

    Xiaolong Wang, Ross Girshick, Abhinav Gupta, and Kaim- ing He. Non-local Neural Networks. In CVPR, pages 7794– 7803, 2018

    2018

  21. [29]

    Flow-Guided Feature Aggregation for Video Object Detection

    Xizhou Zhu, Yujie Wang, Jifeng Dai, Lu Yuan, and Yichen Wei. Flow-Guided Feature Aggregation for Video Object Detection. In ICCV, pages 408–417, 2017

    2017

  22. [30]

    Deep feature flow for video recognition

    Xizhou Zhu, Yuwen Xiong, Jifeng Dai, Lu Yuan, and Yichen Wei. Deep feature flow for video recognition. In CVPR, pages 4141–4150, 2017

    2017

  23. [31]

    Distractor-aware Siamese Networks for Visual Object Tracking

    Zheng Zhu, Qiang Wang, Bo Li, Wei Wu, Junjie Yan, and Weiming Hu. Distractor-aware Siamese Networks for Visual Object Tracking. In ECCV, pages 101–117, 2018

    2018

Pith tools

Reviewed August 14, 2026 · model on record in the stance chip above.