Research map / Video / Video segmentation and tracking
Online Model Distillation for Efficient Video Inference
Ravi Teja Mullapudi, Steven Chen, Keyi Zhang, Deva Ramanan, Kayvon Fatahalian
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 |
|---|---|---|---|
| raise learning rate from 0.01 to 0.1 (ablation) | JITNet 0.8 · LVS 6-video subset | mean IoU: 78.7 -> 16.7 | Hurt |
| lower learning rate from 0.01 to 0.001 (ablation) | JITNet 0.8 · LVS 6-video subset | mean IoU: 78.7 -> 75.6 | Hurt |
| increase max updates per frame from 8 to 16 (ablation) | JITNet 0.8 · LVS 6-video subset | mean IoU: 78.7 -> 78.0 | Hurt |
| reduce max updates per frame from 8 to 4 (ablation) | JITNet 0.8 · LVS 6-video subset | mean IoU: 78.7 -> 77.3 | Hurt |
| increase min stride from 8 to 16 (ablation) | JITNet 0.8 · LVS 6-video subset | mean IoU: 78.7 -> 76.1 | Hurt |
| decrease min stride from 8 to 4 (ablation) | JITNet 0.8 · LVS 6-video subset | mean IoU: 78.7 -> 79.8 | Improved |
| add skip connections between encoder and decoder (ablation) | JITNet 0.8 · LVS 6-video subset | mean IoU: 76.0 -> 78.7 | Improved |
| double network width (ablation) | JITNet 0.8 · LVS 6-video subset | mean IoU: 78.7 -> 80.3 | Improved |
| halve network width (ablation) | JITNet 0.8 · LVS 6-video subset | mean IoU: 78.7 -> 62.0 | Hurt |
| halve input resolution (scale 0.5) (ablation) | JITNet 0.8 · LVS 6-video subset | mean IoU: 78.7 -> 75.9 | Hurt |
| use 75% input resolution (ablation) | JITNet 0.8 · LVS 6-video subset | mean IoU: 78.7 -> 78.1 | No clear effect |
| use MobileNetV2 student (output stride 8) instead of JITNet (ablation) | online distillation · LVS 6-video subset | mean IoU: 78.7 -> 75.3 | Hurt |
Closest papers by meaning
- Delta Distillation for Efficient Video Processing (ECCV 2022 · similarity 0.90)
- Efficient Semantic Video Segmentation with Per-frame Inference (ECCV 2020 · similarity 0.90)
- Learning Efficient Object Detection Models with Knowledge Distillation (NeurIPS 2017 · similarity 0.88)
- AMD: Automatic Multi-step Distillation of Large-scale Vision Models (ECCV 2024 · similarity 0.88)
- Learning Lightweight Object Detectors via Multi-Teacher Progressive Distillation (ICML 2023 · similarity 0.88)
- Dataset Distillation for Pre-Trained Self-Supervised Vision Models (NeurIPS 2025 · similarity 0.88)
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.
