pongsate1 / fashion-parsing Goto Github PK

Repository of my fashion-parsing project. This project is put on hold since I am doing another project now, but will debug if bugs are reported.

Home Page: http://vision.is.tohoku.ac.jp/clothing_parsing

License: Other

CMake 2.87% Makefile 0.65% C++ 78.20% Cuda 5.02% MATLAB 2.22% M 0.01% Python 8.63% HTML 0.51% Shell 0.42% Protocol Buffer 1.48%

python segmentation clothing fashion deep-learning computer-vision caffe

fashion-parsing's Introduction

Fashion-parsing

If you use this work, please cite https://arxiv.org/abs/1703.01386

This work extends fully-convolutional neural networks (FCN) for the clothing parsing problem.

We extend FCN architecture with a side-branch network which we refer outfit encoder to predict a consistent set of clothing labels to encourage combinatorial preference, and with conditional random field (CRF) to explicitly consider coherent label assignment to the given image.

Live demo at http://vision.is.tohoku.ac.jp/clothing_parsing

Project page http://vision.is.tohoku.ac.jp/~tangseng/clothing_parsing_project

Data

Data is in data/. There are three fashion datasets: fashionista-v0.2, fashionista-v1.0, and tmm_dataset_sharing. See the instruction below for data preparation.
Models

Models are in models/. There are 5 models used in fashion parsing: FCN-32s, FCN-16s, FCN-8s, Attribute Layers Training (codename: segc-8s-pre), Attribute Broadcast (codename: sege-8s), and Attribute filtering (codename: attrlog). The folder names are in - format. See the instruction below for training and running the model.
Parsing output and evaluation result

Evaluation results and symbolic links to parsing output are in /public/fashionpose. This folder will be created automatically when run the model. Evaluation results are in json format. The actual output files of Attribute Broadcast (codename: sege-8s), and Attribute filtering (codename: attrlog) model are in the model's folder.
Script

Python script and shell script are in examples/tangseng folder.

Instruction for fashion parsing

Setup following environment: Python, Caffe, and MATLAB
Data preparation Download and convert data into appropiate format according to README and script in each dataset's directory under data/.
Download fcn-32s-pascalcontext.caffemodel according to th url in models/fcn-32s-pascalcontext/readme.md. This model is used as based model for training FCN-32s for fashion datasets.
Train FCN-32s, FCN-16s, FCN-8s, Attribute Layers Training (codename: segc-8s-pre), Attribute Broadcast (codename: sege-8s), and Attribute filtering (codename: attrlog) by execute:

./examples/tangseng/train_all.sh
Run Attribute broadcast (sege) or Attribute filtering (attrlog) network by execute:

./examples/tangseng/run_all.sh

The output will be in models/-/. h5 segmentation output and json evaluation result are expected.
Prepare data for smoothing using CRF by execute:

./examples/tangseng/convert_h5_to_png.sh
Compile CRF by execute:

make -C examples/tangseng/crf
Run CRF smoothing by execute:

./examples/tangseng/run_crf.sh
Run CRF evaluation by execute:

./examples/tangseng/crf_eval.sh
Create symbolic links to output images and refined output images of networks by execute:

./examples/tangseng/createLinkScript.sh

The links are in public/fashionpose/ along with evaluation result in json format. Json files can be open using following command:
```
python -m json.tool <json_file> | less
```

Miscellaneous

I have uploaded my utility library as myutil.py. It contains functions for deprocess an image in h5 files to a regular image for plot, show segmentation maps with colors, etc.

fashion-parsing's People

Contributors

Stargazers

Watchers

Forkers

chenbangfeng labimage asnt huanhuanzhang hrsma2i alice-fen alexlaforge mulinfro pchank nkjzm calvinlcchen rohitn anujonthemove ruchit1395 xtj49 zwcdp joyce725 raghav8360229524

fashion-parsing's Issues

downloading your models?

train_all.sh is taking days (and it crashes due to local issues), is it possible to download your models directly. The whole folder models?
thanks

the result of project

Hi,

Thanks for your project, I have just tested on your live demo and maybe it is not good in some case:

As you can see, the model do not have shoe but it still have it. I hope you can give me some information about this case.

Thanks

quesition about the development environment

Quesition about environment: Python 2.7 or 3.5? the version of MATLAB ,and is Windows 7 ok?

Error when compiling

Hi,

I get this error when I build with make all:

CXX/LD -o .build_release/examples/tangseng/crf/Metric.bin
/usr/lib/gcc/x86_64-linux-gnu/5/../../../x86_64-linux-gnu/Scrt1.o: In function _start': (.text+0x20): undefined reference to main'
collect2: error: ld returned 1 exit status
Makefile:601: recipe for target '.build_release/examples/tangseng/crf/Metric.bin' failed
make: *** [.build_release/examples/tangseng/crf/Metric.bin] Error 1

I tried adding int main () {} in Metric.cpp to continue compiling, but the following error comes up:

CXX examples/tangseng/crf/main.cpp
In file included from examples/tangseng/crf/MaskRefine.h:2:0,
from examples/tangseng/crf/main.cpp:3:
examples/tangseng/crf/FullCRF.h:27:49: error: expected ‘,’ or ‘...’ before ‘&&’ token
PairwisePotential& operator=(PairwisePotential&& other);
^
examples/tangseng/crf/FullCRF.h:53:29: error: expected ‘,’ or ‘...’ before ‘&&’ token
FullCRF& operator=(FullCRF&& other);
^
examples/tangseng/crf/main.cpp: In function ‘int main(int, char**)’:
examples/tangseng/crf/main.cpp:175:18: warning: ignoring return value of ‘int system(const char*)’, declared with attribute warn_unused_result [-Wunused-result]
system("pause");
^
Makefile:552: recipe for target '.build_release/examples/tangseng/crf/main.o' failed
make: *** [.build_release/examples/tangseng/crf/main.o] Error 1

I have no idea why the initial error is happening. Any help would be appreciated.

./examples/tangseng/crf_eval.sh missing

Hi, I am currently working on step 9, but cannot find crf_eval.sh. Is it missing or should be created by any other steps? Thanks.

which version of matlab was used ? Not able to run matlab file on R2018a

your miOU implementation is not correct.

Cannot write to snapshot prefix 'models/fcn-32s-tmm/train'. Make sure that the directory exists and is writeable.

Hi,

Thanks for your project. I have problem with link file:
I1125 15:47:17.463608 19683 hdf5_data_layer.cpp:80] Loading list of HDF5 filenames from: data/tmm_dataset_sharing/TMM_train.h5.txt
I1125 15:47:17.463634 19683 hdf5_data_layer.cpp:94] Number of HDF5 files: 4
F1125 15:47:17.464371 19683 hdf5.cpp:15] Check failed: H5LTfind_dataset(file_id, dataset_name_) Failed to find HDF5 dataset image

I run Matlab on another computer because my computer have problem, so I copy data to my computer again. So do it affect to link dataset???

Thanks

"out of memory" cuda error and required GPU

Hello,

When launching train_all.sh and using a GTX 980Ti 6GB (on Ubuntu 17.10), I encounter a "Check failed: error == cudaSuccess (2 vs. 0) out of memory".
How much GPU memory is required for the training ?

fasionpose datasets

Exceuse me，I‘m doing a research about human fasionpose，and I need the fasionpose datasets，but I can’t connect dataset‘s author，if you hava the dataset，could you please send me a download link？
thx a lot！