Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..
Robust Ego-Exo Correspondence with Long-Term Memory
Authors: Yijun Hu, Bing Fan, Xin Gu, ๆตท้ ไปป, Dongfang Liu, Heng Fan, Libo Zhang
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | In the extensive experiments on the challenging Ego Exo4D benchmark, our method, dubbed LM-EEC, achieves new state-of-the-art results and significantly outperforms existing methods and the SAM 2 baseline, showcasing its strong generalization across diverse scenarios. Our code and model are available at https://github.com/juneyee Hu/LM-EEC. |
| Researcher Affiliation | Academia | 1University of Chinese Academy of Sciences 2University of North Texas 3Institute of Software Chinese Academy of Sciences 4Rochester Institute of Technology Equal contribution and co-first authors Equal advising and corresponding authors |
| Pseudocode | No | The paper describes methods using text, figures (like Figure 2 and 3), and mathematical equations (7-10), but does not present any explicitly labeled 'Pseudocode' or 'Algorithm' blocks. |
| Open Source Code | Yes | Our code and model are available at https://github.com/juneyee Hu/LM-EEC. |
| Open Datasets | Yes | We conduct experiments on the challenging Ego Exo4D benchmark [1]. |
| Dataset Splits | Yes | We adopt the official dataset split in our experiment, where 756 videos are used for training, 202 for validation, and 291 for testing. |
| Hardware Specification | Yes | The model is trained for 60 epochs on 8 NVIDIA A100 GPUs with a batch size of 32 and evaluated on a single V100 GPU, achieving an inference speed of approximately 8.4 FPS. |
| Software Dependencies | No | The paper mentions building upon 'the official SAM 2 base [3]' and using 'Adam W' as an optimizer, but does not provide specific version numbers for these or any other software libraries, frameworks, or programming languages used. |
| Experiment Setup | Yes | Following the training protocol of SAM 2, we sample 8 consecutive frames for each object from ego-exo video pairs, and set the memory bank size to 6. To reduce computational overhead due to the large resolution of original video frames, we resize all frames to 480 480, following the practice in [1]. Given the large-scale nature of Ego Exo4D, we train all modules jointly based on the pre-trained SAM 2 checkpoint, without freezing any components. The model is trained for 60 epochs on 8 NVIDIA A100 GPUs with a batch size of 32 and evaluated on a single V100 GPU, achieving an inference speed of approximately 8.4 FPS. |