About the performance reported in the paper.

[Yun-960]

❓ Questions and Help

Thank you for your work, but in the process of reproducing your code, I cannot achieve the effect reported in the paper.

My training command is as follows: CUDA_VISIBLE_DEVICES=2,3 python -m torch.distributed.launch --master_port 10025 --nproc_per_node=2 tools/relation_train_net.py --config-file "configs/e2e_relation_X_101_32_8_FPN_1x.yaml" MODEL.ROI_RELATION_HEAD.USE_GT_BOX False MODEL.ROI_RELATION_HEAD.USE_GT_OBJECT_LABEL False MODEL.ROI_RELATION_HEAD.PREDICTOR MotifPredictor MODEL.ROI_RELATION_HEAD.CAUSAL.EFFECT_TYPE TDE MODEL.ROI_RELATION_HEAD.CAUSAL.FUSION_TYPE gate SOLVER.IMS_PER_BATCH 12 TEST.IMS_PER_BATCH 2 DTYPE "float16" SOLVER.MAX_ITER 50000 SOLVER.VAL_PERIOD 2000 SOLVER.CHECKPOINT_PERIOD 2000 GLOVE_DIR ./glove MODEL.PRETRAINED_DETECTOR_CKPT "./output/checkpoints/pretrained_faster_rcnn/model_final.pth" OUTPUT_DIR ./output/motifs

My training results:

Hope to get your help. Thank you.

[Yun-960]