Research map / Video / Action recognition
Dynamic Network Quantization for Efficient Video Inference
Ximeng Sun, Rameswar Panda, Chun-Fu (Richard) Chen, Aude Oliva, Rogerio Feris, Kate Saenko
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add knowledge distillation loss from 32-bit model (ablation) | VideoIQ · ActivityNet | mAP: 73.5 -> 74.8 | Improved |
| add knowledge distillation loss from 32-bit model (ablation) | VideoIQ · Mini-Sports1M | mAP: 44.6 -> 46.4 | Improved |
| add efficiency loss L_e (ablation) | VideoIQ · ActivityNet | GFLOPs: 56.4 -> 34.6 | Improved |
| add efficiency loss L_e (ablation) | VideoIQ · ActivityNet | mAP: 75.1 -> 74.5 | Hurt |
| add regularizations L_b and L_d on top of L_e (ablation) | VideoIQ · ActivityNet | GFLOPs: 34.6 -> 28.1 | Improved |
| add L_b regularization alone (ablation) | VideoIQ · ActivityNet | mAP: 74.5 -> 74.3 | Hurt |
| add L_d regularization on top of L_b (ablation) | VideoIQ · ActivityNet | mAP: 74.3 -> 74.8 | Improved |
| add precision choices to skipping (decision space {32,4,2,0} vs {32,0}) (ablation) | VideoIQ · ActivityNet | mAP: 72.9 -> 74.8 | Improved |
| add skipping to precision choices (decision space {32,4,2,0} vs {32,4,2}) (ablation) | VideoIQ · ActivityNet | GFLOPs: 31.4 -> 28.1 | Improved |
| add skipping to precision choices (decision space {32,4,2,0} vs {32,4,2}) (ablation) | VideoIQ · ActivityNet | mAP: 74.5 -> 74.8 | Improved |
| use learned policy instead of random policy (ablation) | VideoIQ · ActivityNet | mAP: 72.8 -> 74.8 | Improved |
| add LSTM to the policy network (ablation) | VideoIQ · ActivityNet | mAP: 74.1 -> 74.8 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 on most shared results.
- AR-Net — AR-Net: Adaptive Frame Resolution for Efficient Action Recognition (ECCV 2020) · better on 4 of 4 shared results
- LiteEval — LiteEval: A Coarse-to-Fine Framework for Resource Efficient Video Recognition (NeurIPS 2019) · better on 4 of 4 shared results
- SCSampler — SCSampler: Sampling Salient Clips From Video for Efficient Action Recognition (ICCV 2019) · better on 4 of 4 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- And the Bit Goes Down: Revisiting the Quantization of Neural Networks (ICLR 2020 · similarity 0.88)
- LEARNED STEP SIZE QUANTIZATION (ICLR 2020 · similarity 0.88)
- Incremental Network Quantization: Towards Lossless CNNs with Low-precision Weights (ICLR 2017 · similarity 0.88)
- Recurrent Residual Module for Fast Inference in Videos (CVPR 2018 · similarity 0.88)
- DSConv: Efficient Convolution Operator (ICCV 2019 · similarity 0.88)
- Quantization without Tears (CVPR 2025 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
