bar371 / geff Goto Github PK

Thank you for the hardwork and sharing of your code, I have encountered some issues while attempting to replicate the results for the PRCC dataset with the CTL model.
I executed the inference script:
python /GEFF/Scripts/inference.py prcc CTL --reid_config /GEFF/ReIDModules/centroids_reid/configs/256_resnet50.yml --dataset_path /GEFF/datasets/prcc --device "cuda:0" --reid_checkpoint GEFF/checkpoints/CTL/dukemtmcreid_resnet50_256_128_epoch_120.ckpt --alpha 0.75 --detection_threshold 0.7
Actual Results:
Setting seed to: 0
Traceback (most recent call last):
File "/GEFF/Scripts/inference.py", line 472, in
main(args)
File "/GEFF/Scripts/inference.py", line 467, in main
run_inference(args, config=config)
File "/GEFF/Scripts/inference.py", line 399, in run_inference
qd_pids, qd_camids, qd_clothes_ids, g_pids, g_camids, g_clothes_ids = reid_inference_on_prcc(args, config)
File "/GEFF/Scripts/inference.py", line 184, in reid_inference_on_prcc
reid_model.init_model(config, args.reid_checkpoint)
File "/GEFF/./ReIDModules/CTL_model.py", line 17, in init_model
self.model = CTL._load_model_state(checkpoint)
AttributeError: type object 'CTLModel' has no attribute '_load_model_state'

I have double-checked the steps and parameters, but I still face difficulties .Any guidance on how to proceed or troubleshoot this issue would be greatly appreciated.

Hello @bar371 ! Thank you for the hardwork and sharing of your code. I would like to know how we can run the script on our own video data.

Hi, great work! I cannnot reproduct your prcc result by the following setting. Please teach me～

python Scripts/inference.py "prcc" "CAL" --reid_config ReIDModules/CAL/configs/res50_cels_cal.yaml --dataset_path "prcc" --device "cuda:0" --reid_checkpoint "prcc-checkpoint.pth.tar" --alpha 0 --detection_threshold 0.7

・PRCC Face Module without Enriched Gallery
Computing CMC and mAP Computing CMC and mAP for the same clothes setting
Results ---------------------------------------------------
top1:28.9% top5:41.4% top10:44.8% top20:50.3% mAP:18.0%

Computing CMC and mAP Computing CMC and mAP only for clothes changing
Results ---------------------------------------------------
top1:13.8% top5:17.9% top10:19.7% top20:21.2% mAP:8.0%

Hello, could you please tell me how long it took you to run gallery_enrichment.py on CCIVD? I've been running it on a cloud server with a 16GB DPU for almost two days, and the code is still not finished. Is this situation normal?

Hello @bar371 ! Thank you for the hardwork and sharing of your code.

My question is regarding of the project code:
Is it possible for the project code to predict labels for unlabeled query images after the gallery has been created?
In my understanding, once the gallery is set up with labeled images, I would like to know if the system can infer and suggest labels for new, unlabeled query images based on the gallery's dataset.

I am looking forward to your guidance on this matter. Thank you for your valuable contributions to the field and for considering my question.

Description

I have encountered some issues while attempting to replicate the results using the provided Colab notebook steps for the PRCC dataset with the CAL model.The environment I use is Kaggle Notebook.

Steps to Reproduce

1. I set the parameters as instructed in the Colab notebook:

   REID_MODEL = 'CAL'  # Using the CAL model
   DATASET = 'prcc'  # Specifying the PRCC dataset
   DATASET_PATH = '/kaggle/working/prcc'
   DEVICE = 'cuda:0'  
   REID_CHECKPOINT = '/kaggle/input/geff-code/GEFF/GEFF-main/CAL_checkpoints'    
   DETECTION_THRESHOLD_BY_DATASET = {'ltcc': 0.8, 'prcc': 0.7, 'ccvid': 0.5, 'last': 0.7, 'vcclothes': 0.8}
   SIMILARITY_THRESHOLD_BY_DATASET = {'ltcc': 0.5, 'prcc': 0.65, 'ccvid': 0.75, 'last': 0.45, 'vcclothes': 0.5}
   DETECTION_THRESHOLD = DETECTION_THRESHOLD_BY_DATASET[DATASET]
   SIMILARITY_THRESHOLD = SIMILARITY_THRESHOLD_BY_DATASET[DATASET]

2. I ran the command to build the enriched gallery:

!python Scripts/gallery_enrichment.py $DATASET --dataset_path $DATASET_PATH --detection_threshold $DETECTION_THRESHOLD --similarity_threshold $SIMILARITY_THRESHOLD --device $DEVICE

3. Finally, I executed the inference script:

!python Scripts/inference.py $DATASET $REID_MODEL --reid_config /kaggle/input/geff-code/GEFF/GEFF-main/ReIDModules/CAL/configs/res50_cels_cal.yaml --dataset_path $DATASET_PATH --device $DEVICE --reid_checkpoint /kaggle/input/geff-code/GEFF/GEFF-main/CAL_checkpoints/CAL/prcc-checkpoint.pth.tar --alpha 0.75 --detection_threshold $DETECTION_THRESHOLD

Actual Results

I have double-checked the steps and parameters, but I still face difficulties in achieving the expected outcome.
Any guidance on how to proceed or troubleshoot this issue would be greatly appreciated.

Thank you for the hardwork and sharing of your code. "In the supplementary material, we examine different α values.", could you please tell me where I can find this file?

Great work! I would like to suggest adding a code running example in the README, using either the PRCC or LTCC dataset as an example. Thank you."

I'm very interested in your work and I look forward to the release of the code