CCTrans: Simplifying and Improving Crowd Counting with Transformer(Code reproduction)

• Code reproduction
• Original paper Link

Overview

• Presentate only the experiment on dataset ShanghaiTech Part A (loss: DM-Count)
• ShanghaiTech Part A

Code	MAE	MSE
PAPER	54.8	86.6
This code	54.20	88.97

Our code reaches this result with the standard hyperparameter set in code. Trained with batch-size=8 for around 1500 epoch(as said in the paper). Best validation at around epoch 606

code framework

• adopt code of DM-Count.
• link

Pre-trained weights

• Download pretrained weights for alt_gvt_large.pth link. Move the file under model_weights folder

Training

Take a look at the arguments accepted by train.py

• update root "data-dir" in ./train.py.
• [new] Added wandb integration. If you want to log with wandb, set --wandb 1 in train.py after having logged in to wandb (wandb login in console)
• launch with python train.py

Testing

• python test_image_patch.py
• Due to crop training with size of 256x256, the validation image is divided into several patches with size of 256x256, and the overlapping area is averaged.
• Download the pretrained model from Baidu-Disk(提取码: se59) link

Visualization

• python vis_densityMap.py
• save to ./vis/part_A_final

TensorRT

Require torch2trt

See torch_to_trt.py

Environment

See requirements.txt