ternaus/cloths_segmentation

Code for binary segmentation of cloths

Code for binary segmentation of various cloths

Installation

pip install -U cloths_segmentation

Example inference

Jupyter notebook with the example:

WebApp

https://clothssegmentation.herokuapp.com/

Data Preparation

Download the dataset from https://www.kaggle.com/c/imaterialist-fashion-2019-FGVC6

Process the data using script

The script will create process the data and store images to folder images and binary masks to folder labels.

Training

Define the config.

Example at cloths_segmentation/configs

You can enable / disable datasets that are used for training and validation.

Define the environmental variable IMAGE_PATH that points to the folder with images.

Example:

export IMAGE_PATH=<path to the the folder with images>

Define the environmental variable LABEL_PATH that points to the folder with masks.

Example:

export MASK_PATH=<path to the folder with masks>

Training

python -m cloths_segmentation.train -c <path to config>

Inference

python -m torch.distributed.launch --nproc_per_node=<num_gpu> cloths_segmentation/inference.py \
                                   -i <path to images> \
                                   -c <path to config> \
                                   -w <path to weights> \
                                   -o <output-path> \
                                   --fp16

cloths_segmentation's Issues

Issues with starting Training

I have set up all the required data and the paths but still get this error all the configs

Traceback (most recent call last):
File "/usr/lib/python3.9/runpy.py", line 197, in _run_module_as_main
return _run_code(code, main_globals, None,
File "/usr/lib/python3.9/runpy.py", line 87, in _run_code
exec(code, run_globals)
File "/home/electron/cloths_segmentation/cloths_segmentation/train.py", line 189, in
main()
File "/home/electron/cloths_segmentation/cloths_segmentation/train.py", line 175, in main
pipeline = SegmentPeople(hparams)
File "/home/electron/cloths_segmentation/cloths_segmentation/train.py", line 46, in init
state_dict = state_dict_from_disk(
File "/home/electron/.local/lib/python3.9/site-packages/iglovikov_helper_functions/dl/pytorch/utils.py", line 32, in state_dict_from_disk
checkpoint = torch.load(file_path, map_location=lambda storage, loc: storage)
File "/home/electron/.local/lib/python3.9/site-packages/torch/serialization.py", line 594, in load
with _open_file_like(f, 'rb') as opened_file:
File "/home/electron/.local/lib/python3.9/site-packages/torch/serialization.py", line 230, in _open_file_like
return _open_file(name_or_buffer, mode)
File "/home/electron/.local/lib/python3.9/site-packages/torch/serialization.py", line 211, in init
super(_open_file, self).init(open(name, mode))
FileNotFoundError: [Errno 2] No such file or directory: '2020-10-29/epoch=4.ckpt'

Segment only one cloth per time

Hello, thank you for this great work.
I just want to know if is it possible to use this model to get only one cloth per time, like getting only pants from a full body photo

load

could you give some samples that load a model from a local directory instead of download from url? Thanks a lot

timm

whihc version

----> 1 from timm.models import ByoModelCfg, ByoBlockCfg, ByobNet
2
3 from ._base import EncoderMixin
4 import torch.nn as nn
5

ImportError: cannot import name 'ByoModelCfg'

post processing

thank you for this amazing repo. the model works really well.
I have one query, are you using some sort of heuristics/post-processing for the demo. The model output and the web demo output do not match in many images (web app work very well actually compared to the pre-trained model).

ImportError: cannot import name 'ByoModelCfg'

from cloths_segmentation.pre_trained_models import create_model

----> 1 from timm.models import ByoModelCfg, ByoBlockCfg, ByobNet
2
3 from ._base import EncoderMixin
4 import torch.nn as nn
5

ImportError: cannot import name 'ByoModelCfg'

Save model

Hello
Your model is incredible, but I have a problem, I always have to train my own model because I didn't see any save model in your train script. How do you save the model?
Thanks so much