heitorrapela / HalluciDet

[WACV2024] HalluciDet: Hallucinating RGB Modality for Person Detection Through Privileged Information (Accepted at WACV 2024 and LatinX@CVPR2024 Extended Abstract)
MIT License
19 stars 0 forks source link

pretrained weights #2

Closed medicineqian closed 2 months ago

medicineqian commented 4 months ago

Hi, can you provide the test code and pretrained weights?

heitorrapela commented 4 months ago

Hi, thanks for your interest in our work. The training script in the main repo can also be used for test, but I will provide an easier way to do it in the next update. For the weights, I will also provide easy access to it, probably in the next few days I will share here, Best Regards, Heitor Medeiros

heitorrapela commented 2 months ago

Hi, sorry for the delay, we updated the code, now we have the eval_hallucidet.py script, and some pre-trained weights https://huggingface.co/heitorrapela/hallucidet/tree/main , to eval hallucidet, you have to download the detector and hallucidet weights.

After you download, you can place the weights inside ./checkpoints/ and run the following command:

CUDA_VISIBLE_DEVICES=0 python eval_hallucidet.py --pretrained --modality ir --detector-path ./checkpoints/llvip/seed123/fasterrcnn_rgb_llvip_seed123.ckpt --hallucidet-path ./checkpoints/llvip/seed123/hallucidet_llvip_seed123.ckpt --wandb-project wacv2024 --wandb-name detector_fasterrcnn_hallucidet_det01reg01_llvip_200ep_seed123 --detector fasterrcnn --dataset llvip --epochs 1 --batch 8 --seed 123