dearcaat/pict

pytorch implementation of PicT

PicT

This repo is the official implementation of "PicT: A Slim Weakly Supervised Vision Transformer for Pavement Distress Classification" based on Pytorch.

For more details of the pavement dataset CQU-BPDD used in paper, please refer to CQU-BPDD. (Note: CQU-BPDD can be only used in the uncommercial case and is licensed under CC BY-NC-SA 4.0.)

For more details of this task, see Pavement Distress Classification.

Requirement

Core:

• torch == 1.11+cu11.5 (Not required, but recommended.)
• timm == 0.67
• ...

Docker:

Download Docker image via Baidu Cloud.

Note: If u wanto reproduce the results in the paper exactly, please email Wenhao.

Usage

These examples are in the I-REC setting. For other settings, please change the config file.

Train Swin-S

python3 main.py --data-path=$DATA_PATH --output=$OUTPUT_PATH --project=pict --cfg ../configs/baseline/swin_small_1rec.yaml --title=swin_s

Train PicT

# PicT uses the pretrained swin_s weight to init the teacher model
python3 main.py --data-path=$DATA_PATH --output=$OUTPUT_PATH --project=pict --cfg ../configs/baseline/swin_small_1rec.yaml ../configs/pict_1rec.yaml --title=pict --opt PICT.TEACHER_INIT $PRETRAINED_WEIGHT_PATH

pict's Issues

Error in training with DDP mode

When I train PicT with DDP mode, I find this error:
File "main.py", line 207, in main
models['main'] = NativeDDP(models['main'], device_ids=[config.LOCAL_RANK ]) # can use device str in Torch >= 1.1models['main'] = NativeDDP(models['main'], device_ids=[config.LOCAL_RANK ]) # can use device str in Torch >= 1.1
File "/eto_cn1/miniconda3/envs/XXX/lib/python3.8/site-packages/torch/distributed/utils.py", line 89, in _verify_param_shape_across_processes
return dist._verify_params_across_processes(process_group, tensors, logger)
RuntimeError: NCCL error in: /opt/conda/conda-bld/pytorch_1659484657607/work/torch/csrc/distributed/c10d/ProcessGroupNCCL.cpp:1191, invalid usage, NCCL version 2.10.3
ncclInvalidUsage: This usually reflects invalid usage of NCCL library (such as too many async ops, too many collectives at once, mixing streams in a group, etc).
return dist._verify_params_across_processes(process_group, tensors, logger)return dist._verify_params_across_processes(process_group, tensors, logger)

Wish to get your reply