Research map / Multi-agent systems and games / Human-AI and goal-conditioned
Learning Precise Affordances from Egocentric Videos for Robotic Manipulation
Gen Li, Nikolaos Tsagkas, Jifei Song, Ruaridh Mon-Williams, Sethu Vijayakumar, Kun Shao, Laura Sevilla-Lara
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 |
|---|---|---|---|
| replace DeiT III backbone with DINOv2 (ablation) | baseline affordance segmenter · AED (ablation) | mIoU: 31.02 -> 45.45 | Improved |
| add embedder MLP before classification (ablation) | DINOv2 segmenter · AED (ablation) | mIoU: 45.45 -> 48.83 | Improved |
| add 4x upsampling after embedder (ablation) | DINOv2 segmenter · AED (ablation) | mIoU: 48.83 -> 51.41 | Improved |
| add focal and dice loss instead of BCE (ablation) | DINOv2 segmenter · AED (ablation) | mIoU: 50.70 -> 53.12 | Improved |
| use cosine similarity classifier instead of linear layer (ablation) | DINOv2 segmenter · AED (ablation) | mIoU: 54.96 -> 56.70 | Improved |
| remove learned background classifier (implicit background via threshold) (ablation) | cosine similarity segmenter · AED (ablation) | mIoU: 55.52 -> 56.70 | Improved |
| add Depth Feature Injector used only in training (ablation) | affordance segmenter · AED (ablation) | mIoU: 56.70 -> 60.15 | Improved |
| use DFI also at inference (ablation) | affordance segmenter · AED (ablation) | mIoU: 60.15 -> 64.66 | Improved |
| add LoRA fine-tuning of DINOv2 (ablation) | segmenter with DFI · AED (ablation) | mIoU: 64.66 -> 68.62 | Improved |
| discard DFI at inference (ablation) | segmenter with DFI · AED inference efficiency | inference time (ms): 10.1 -> 6.3 | Improved |
| use DINOv2 foundation backbone instead of ImageNet-pretrained SegFormer | segmentation baselines · AED (721 images, zero-shot) | mIoU: 23.72 -> 46.16 | Improved |
Closest papers by meaning
- RAGNet: Large-scale Reasoning-based Affordance Segmentation Benchmark towards General Grasping (ICCV 2025 · similarity 0.94)
- A0: An Affordance-Aware Hierarchical Model for General Robotic Manipulation (ICCV 2025 · similarity 0.93)
- AffordDP: Generalizable Diffusion Policy with Transferable Affordance (CVPR 2025 · similarity 0.92)
- VidBot: Learning Generalizable 3D Actions from In-the-Wild 2D Human Videos for Zero-Shot Robotic Manipulation (CVPR 2025 · similarity 0.91)
- SeqAfford: Sequential 3D Affordance Reasoning via Multimodal Large Language Model (CVPR 2025 · similarity 0.90)
- Weakly Supervised Multimodal Affordance Grounding for Egocentric Images (AAAI 2024 · similarity 0.90)
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.
