Very poor result when evaluate on Pascal VOC dataset

[Kent Vu]

Thanks for your's excellent paper and repository!
But I have a problem when I try to evaluate on some datasets. I follow instructions in this #46
I try to evaluate the pre-trained RegionClip on Pascal VOC by using this command

• To create content embedding
  !python3 ./tools/extract_concept_features.py \ --config-file ./configs/LVISv1-InstanceSegmentation/CLIP_fast_rcnn_R_50_C4_zsinf.yaml \ MODEL.WEIGHTS ./pretrained_ckpt/regionclip/regionclip_pretrained-cc_rn50.pth \ MODEL.CLIP.OFFLINE_RPN_CONFIG ./configs/LVISv1-InstanceSegmentation/mask_rcnn_R_50_FPN_1x.yaml \ INPUT_DIR ./datasets/custom_concepts \ OUTPUT_DIR ./output/concept_feats \ MODEL.CLIP.GET_CONCEPT_EMB True
• Using this to evaluate
  !python3 ./tools/train_net.py \ --eval-only \ --num-gpus 1 \ --config-file ./configs/PascalVOC-Detection/faster_rcnn_R_50_C4.yaml \ MODEL.WEIGHTS ./pretrained_ckpt/regionclip/regionclip_pretrained-cc_rn50.pth \ MODEL.CLIP.TEXT_EMB_PATH ./output/concept_feats/concept_embeds.pth \ MODEL.CLIP.CROP_REGION_TYPE GT \ MODEL.CLIP.MULTIPLY_RPN_SCORE False

BUT I get very poor results. Can you help me with this

[Nguyet Nguyen]

@kent252