Research map / Label-efficient and robust learning / Domain adaptation and generalization
VisCoP: Visual Probing for Video Domain Adaptation of Vision Language Models
Dominick Reilly, Manish Govind, Le Xue, Srijan Das
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 visual probes to a frozen vision encoder (VisCoP) instead of finetuning VE | VLM (VideoLLaMA3) · egocentric target | delta target vs base: +0.57 (full finetune) -> +3.53 | Improved |
| finetune connector, vision encoder and LLM (ablation) | VLM · egocentric source | delta source: -0.50 | Hurt |
| finetune vision encoder with connector for depth adaptation (ablation) | VLM · depth source | delta source: -3.87 | Hurt |
| add VisCoP probes for depth adaptation | VLM · depth target | delta target: +16.44 (connector only) -> +19.27 | Improved |
| train vision encoder in robot control adaptation (ablation) | VLM · robot control source | delta source: -8.87 (VE+LLM) vs -4.58 (VisCoP) | Hurt |
| add VisCoP in robot control simulation | VLM · VIMA-Bench | avg success rate: 65.13 -> 67.82 | Improved |
| add VisCoP in real-world robot with xArm-Det joint training | VLM · xArm real-world | avg success rate: 80.00 -> 96.67 | Improved |
| restrict probe interaction to spatial only (ablation) | VisCoP · egocentric | delta target: +3.53 -> +1.83 | Hurt |
| restrict probe interaction to temporal only (ablation) | VisCoP · egocentric | delta target: +3.53 -> -0.30 | Hurt |
| use KL regularization toward base VLM on depth adaptation (ablation) | connector+VE finetuning · depth | source avg: 68.9 -> 73.0 | Improved |
| add source replay (up to 25%) on depth adaptation (ablation) | connector+VE finetuning · depth | source avg: 68.9 -> 70.0 | Mixed |
| increase number of probes beyond 16 (ablation) | VisCoP · three domains | delta target / source | No clear effect |
Closest papers by meaning
- CoVFT: Context-aware Visual Fine-tuning for Multimodal Large Language Models (CVPR 2026 · similarity 0.89)
- Frustratingly Easy Test-Time Adaptation of Vision-Language Models (NeurIPS 2024 · similarity 0.89)
- ViSurf: Visual Supervised-and-Reinforcement Fine-Tuning for Large Vision-and-Language Models (ICML 2026 · similarity 0.89)
- Synthesize Diagnose and Optimize: Towards Fine-Grained Vision-Language Understanding (CVPR 2024 · similarity 0.89)
- VISion On Request: Enhanced VLLM efficiency with sparse, dynamically selected, vision-language interactions (CVPR 2026 · similarity 0.89)
- AdaptVision: Efficient Vision-Language Models via Adaptive Visual Acquisition (CVPR 2026 · similarity 0.89)
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.
