FCT icon indicating copy to clipboard operation
FCT copied to clipboard

Code for CVPR 2022 Oral paper: 'Few-Shot Object Detection with Fully Cross-Transformer'

Results 21 FCT issues
Sort by recently updated
recently updated
newest added

Hello author, I have seen how to use the previous, but I still do not understand how to insert in the code, especially about idx_ and head_ how to set

I simply add it near ‘attn_x’ and set 'query_image' the picture I predict, but how to set the 'support_image' ?

File "/data/sam/FCT-main/faster_rcnn_train_net.py", line 85, in main return trainer.train() File "/data/anaconda3/envs/py37_dr2/lib/python3.7/site-packages/detectron2/engine/defaults.py", line 484, in train super().train(self.start_iter, self.max_iter) File "/data/anaconda3/envs/py37_dr2/lib/python3.7/site-packages/detectron2/engine/train_loop.py", line 149, in train self.run_step() File "/data/anaconda3/envs/py37_dr2/lib/python3.7/site-packages/detectron2/engine/defaults.py", line 494, in run_step self._trainer.run_step()...

I have a problem when I perform "sh scripts/single_branch_pretraining_coco_pvt_v2_b2_li.sh" . ![Capture001](https://github.com/GuangxingHan/FCT/assets/151814406/4bf3c239-0098-4a9b-9e47-614267ce01f7) And I added "WEIGHT_DECAY_BIAS: 0.0001" in the config file single_branch_pretraining_coco_pvt_v2_b2_li.yaml,now it runs successfully. I was wondering if "WEIGHT_DECAY_BIAS" should...

I've met some problems when I trained pascalvoc datasets.I run the scripts/single_branch_pretraining_pascalvoc_split1_pvt_v2_b2_li.sh file is ok. But ,when I run scripts/two_branch_training_pascalvoc_split1_pvt_v2_b2_li.sh and scripts/two_branch_few_shot_finetuning_pascalvoc_split1_pvt_v2_b2_li.sh files, it said i don't have './datasets/pascal_voc/voc_2007_trainval_all1_10shot.pkl' file...

I cannot find the dictory 'new_annotations/final_split_voc_1_shot_instances_train2014 .json'.How can I generate it?I don't find the script to generate it.

Great work done!, How can I replace the pvt backbone to swin transformer backbone? Is there a common standard for modifying the backbone in detectron2?

Your code takes more than 5 hours under test conditions, can you provide a GPU test version self._cpu_device = torch.device("cuda" if torch.cuda.is_available() else "cpu") replace self._cpu_device = torch.device("cpu") it isn't...