A multimodal RGB-depth fusion backbone with vision transformer, masked-token contrastive learning, and curriculum domain randomization outperforms baselines in simulation and enables zero-shot real-world robot manipulation.
Grad-CAM: Visual Explanations from Deep Networks via Gradient-Based Localization
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2025 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Multimodal Fusion for Sim2real Transfer in Visual Reinforcement Learning
A multimodal RGB-depth fusion backbone with vision transformer, masked-token contrastive learning, and curriculum domain randomization outperforms baselines in simulation and enables zero-shot real-world robot manipulation.