Hi, thanks for releasing the code and data! I’m trying to reproduce the scene-graph results, but the repository only provides the COCO initialization weights. Could you please share the fine-tuned checkpoints used in the paper (e.g., the ones corresponding to the best bbox mAP)? Having those would let us verify the benchmark numbers and serve as a stronger starting point for downstream experiments. Thanks a lot!
Hi, thanks for releasing the code and data! I’m trying to reproduce the scene-graph results, but the repository only provides the COCO initialization weights. Could you please share the fine-tuned checkpoints used in the paper (e.g., the ones corresponding to the best bbox mAP)? Having those would let us verify the benchmark numbers and serve as a stronger starting point for downstream experiments. Thanks a lot!