xthan / viton Goto Github PK
View Code? Open in Web Editor NEWCode and dataset for paper "VITON: An Image-based Virtual Try-on Network"
Code and dataset for paper "VITON: An Image-based Virtual Try-on Network"
What configurations, aspect ratios & paddings did you use to generate segmentation maps?
LIP Human Parser crops images to 640X640. So if I use my 762X1100 image it crops it to 640. Now, if I use a resized image(640X640) for segmentation & use 762X1100 as the test image. I think the segmentation map generated on 640X640 doesn't maps well onto 762X1100 test image.
Please explain what padding & aspect ratios did you use for training & testing and segmentation map generation
Thank you
Hi, I test your stage2 model with test_stage2.sh, some results are good, but some are all black.
Do you know possible problem? Thank you!
I know that that "the dataset is no longer publicly available", however I wanted to get access to the data to try code. How can I get access to the dataset?
I try to run your code but I am getting this error:
Traceback (most recent call last):
File "model_zalando_mask_content_test.py", line 236, in
tf.app.run()
File "/BS/wild-search-gaze2/work/3D_body_shape/SMPLify/SMP_env/local/lib/python2.7/site-packages/tensorflow/python/platform/app.py", line 124, in run
_sys.exit(main(argv))
File "model_zalando_mask_content_test.py", line 136, in main
prod_segment_holder, image_holder)
File "/BS/wild-search-gaze2/archive00/VITON-master/model_zalando_mask_content.py", line 200, in create_model
vgg_real = build_vgg19(image, FLAGS.vgg_model_path)
File "/BS/wild-search-gaze2/archive00/VITON-master/utils.py", line 482, in build_vgg19
vgg_rawnet = scipy.io.loadmat(model_path)
File "/BS/wild-search-gaze2/work/3D_body_shape/SMPLify/SMP_env/local/lib/python2.7/site-packages/scipy/io/matlab/mio.py", line 141, in loadmat
MR, file_opened = mat_reader_factory(file_name, appendmat, **kwargs)
File "/BS/wild-search-gaze2/work/3D_body_shape/SMPLify/SMP_env/local/lib/python2.7/site-packages/scipy/io/matlab/mio.py", line 64, in mat_reader_factory
byte_stream, file_opened = _open_file(file_name, appendmat)
TypeError: 'NoneType' object is not iterable
I realise the dataset isn't available for free use now, but would it be possible to shed light on what kind of data was used?
Hi, I found that in your 'model_zalando_tps_warp.py'
in the create_model() function, you will do "create_generator" again, which has done in stage1.
So, this code ('model_zalando_tps_warp.py') will train stage1 and then stage2, rather than train only stage2?
Thanks a lot again!
I have trained the model on my data set. While testing(stage1) it's assigning same color product to all the result images. When I change the mode to train in test_stage1.sh, it is giving good results. Was there anything wrong with my training?
Hello!
I 'm now trying to run preprocess_viton.sh to generate TF records. But it is hanging now with a TypeError saying "a bytes-like" object us required, not 'str' (shown in following screenshot). Is there anything wrong with my Python? I am new to Python. I will be appreciate if someone helps solve this problem~
I am trying to inference on pictures beside the pictures that is on dataset right now. the problem is that the human parsing segmentation that is used in dataset have 14 classes but the lip dataset and github repository that you introduced for human parsing have 20 classes.
anyone knows, in dataset, which network is used for human parsing segmentation?
Hi!
Thanks for your project!
I wanted to train a network with my own data, that's why I had to use network from Realtime Multi-Person Pose Estimation repository to prepare data.
But in pose.pkl there are dictionaries with "candidate" and "subset" keys and I don't see them in OpenPose output. Maybe it's possible to match them?
hi,
I also met the problem : the first train stage hangs and just shows the infomation:
I use the parameters:
--output_dir "model/stage1/"
--input_file_pattern "./prepare_data/tfrecord/zalando-train-?????-of-00032"
I have read : https://github.com/xthan/VITON/issues/2
I test tensorflow 1 & 4 & 5 but it still hangs and shows:
INFO:tensorflow:Starting standard services.
INFO:tensorflow:parameter_count = 29269632
INFO:tensorflow:Starting queue runners.
and then nothing happen.
I want to know about the required GPU capacity and the required environment.
Any ideas ?
Thanks.
I'm having problem getting into
https://drive.google.com/drive/folders/1qFU4KmvnEr4CwEFXQZS_6Ebw5dPJAE21?usp=sharing
Is that because of mine internet setting?Can you check it please?
Getting Keyerror: segment.
segment_raw = sio.loadmat(os.path.join(
tf.app.flags.FLAGS.segment_dir, image_id))['segment'] #getting the error on this line
segment_raw = process_segment_map(segment_raw, image.shape[0], image.shape[1])
pose_raw = sio.loadmat(os.path.join(FLAGS.pose_dir, image_id))
What would have been the issue and how to resolve it:?
Getting error in test stage1.
loading model from checkpoint model/stage1/model-15000 INFO:tensorflow:Restoring parameters from model/stage1/model-15000 Traceback (most recent call last): File "model_zalando_mask_content_test.py", line 236, in <module> tf.app.run() File "/home/gamut/.conda/envs/tf_3/lib/python3.6/site-packages/tensorflow/python/platform/app.py", line 125, in run _sys.exit(main(argv)) File "model_zalando_mask_content_test.py", line 156, in main test_info = open(FLAGS.test_label).read().splitlines() FileNotFoundError: [Errno 2] No such file or directory: 'data/viton_test_pairs.txt'
I have dowloaded the models and moved them in model folder as mentioned.
How to move forward now?
Hi, I am trying to test the Model. I have both of the Person Segmentation and Pose Estimation setup, but I am unsure of what I need to save and use as input to Stage 1 of VITON. In prepare_data/build_viton.py, it is looking for a pickle file. But for testing, it does not specify what file type it is looking for..
I understand that it is not available for commercial use, but we would like to conduct research strictly for academic purposes and thesis work. How can we do this?
Hi, I'm trying to make your model works with other types of clothing like pants, skirt.
But none of my ideas seem to work. Can you give me some ideas of how to make your model more generalize? Thanks
Can you provide some tips on what sites you chose to scrape the training data?
Thank you very much
1.According to your paper,your coarse network takes no rgb information other than the rgb segmentation of the women's hair and head.
I'm wondering how does the network manage to 'guess' the color of the pants and the shape and color the the woman's arm ,when no information tells the network that the color of the pants is gray?
Is this a sign of overfitting?
2.If you can already get the segmentation of a person using the LIP, why bother to train the network and get the mask yourself?
3.When I implemented LIP,I found that its segmentation results are not very satisfying, far from being able to create an accurate mask of the clothes or hair of head? Do you have similar problems?
Thanks a lot.
Hi, I want to train this work on my own data, and when I run extract_tps.m, I always get ans='not enough'. Also there is no folder ’data/tps/‘. I don't know why. Looking forward to hearing from you. Thank you so much.
Excuse me , I want to test my dataset to this but I don't know how to generate the pose file.
I went to see the document on README but i still don't know what is subset.
The candidate is the x,y of the keypoint, right?
What is the subset ?
leaky relu work great compare to relu , but in vition stage1 network for encoder activation function leaky rely and for decoder expect last layer activation relu.
can I know reason why relu for decoder network not leaky relu.
I have googled a lot , but didnot find any good explanation
Hello,
I am running this code.
Stage 1 works fine.
Stage 2 runs but outputs black images. I am using Octave and not Matlab to run shape_context_warp.m. Is the output of this available online?
Regards,
Daniel
I found the segmentation result of person image sometimes is annoying. For example, for the upper clothes, there may output several different labels(such as upper_clothes, dress, skirt etc.) As a result, it's hard to decide to take which part as the real upper clothes region.
Could you please show the code for processing this kind of cases? Thanks a lot.
I am trying to make a custom dataset identical to the VITON data for training. I have tried running the densepose which outputs two images: one IUV and one INDS image. However, As test rn i compared both the outputs by converting them to numpy files with the numpy files in the train_densepose folder. The outputs don't match exactly. Also, the shape is the same but the original numpy file is three rows shorter. Can you kindly guide me at what settings did you run the densepose to get the required .npy files included in train_densepose folder.
Hi,
I tried training the first stage using "model_zalando_mask_content.py", and set the following parameters:
--output_dir "model/stage1/"
--input_file_pattern "./prepare_data/tfrecord/zalando-train-?????-of-00032"
However, the training hangs when executing the following line:
results = sess.run(fetches, options=options, run_metadata=run_metadata)
Any ideas?
Thanks a lot!
hi, how can I create dataset located in data folder like pose, segment, ... I want to train with my images. thank you
I'm trying to generate tps files in data/tps for my data by running extract_tps in Matlab. But after generating for few images it is hanging.
The stage one output is totally weird. I did run teststage1. I have attached a samples output for test stage 1.
I also did modify line 52 from tf.flags.DEFINE_integer("end", "2032", "") to tf.flags.DEFINE_integer("end", "1", "") or else the loop was trying to run for 2032 times and i got this error.
File "model_zalando_mask_content_test.py", line 161, in main
info = test_info[j].split()
IndexError: list index out of range
Can someone please help me with this?
I make data/vision_test_pairs.txt like this,
001_0.jpg 001_1.jpg
When I run test_stage1.sh, I get this error. I'm sorry I don't know how to move forward. Are there some examples of pictures could share with me. I don't know how to fix the data of pose, segment and women_top. Thank you so much.
I am trying to train models using my dataset .
After compiling extract_tps( https://github.com/xthan/VITON/blob/master/prepare_data/extract_tps.m ) i got this error at line number 32 .
warning: textread is obsolete; use textscan instead
h = 128
w = 96
ans = 1
error: V2(_,481): out of bound 256 (dimensions are 192x256)
error: called from
extract_tps at line 32 column 10
I got this error even if my data set images size is 192 * 256(width * height)
could you please help me to resolve it.
Hi, I just want to ask question about how can keep background with wild image. I try with wild image but background disapear
The input images are resized to 256*192, is it to fit the VGGnet that the height is set to 256 ?
Have you tried the larger input size to train the network? Will the larger size conduct worse results?
Looking forward to your reply,thanks.
@xthan
For different clothing materials, such as linen clothes and a clothing figure wearing cotton, can this project be realized, if it is different fat and thin people?
is it possible to provide an example test pair of images and their mat files+pose? I.e. everything needed to test the test part of the code? to make sure all works. even synthetic or toy will work.
hey!
this is awesome! how's the implementation for the other stages coming along? if it's not on the roadmap i'm happy to have a go.
in vgg19 loss your used some value ,I was wonder how did these values are taken , I gone to your paper , but there is no information regarding this
p1 = compute_error(vgg_real['conv1_2'],
vgg_fake['conv1_2']) / 5.3 * 2.5 # 12812864
p2 = compute_error(vgg_real['conv2_2'],
vgg_fake['conv2_2']) / 2.7 / 1.2 # 6464128
p3 = compute_error(vgg_real['conv3_2'],
vgg_fake['conv3_2']) / 1.35 / 2.3 # 3232256
p4 = compute_error(vgg_real['conv4_2'],
vgg_fake['conv4_2']) / 0.67 / 8.2 # 1616512
p5 = compute_error(vgg_real['conv5_2'],
vgg_fake['conv5_2']) / 0.16 # 88512
can your help about how did got this value 2.5 ,1.2 , 2.3 , 8.2 , p1-p4 ?
Hi, thank you for your work. Could you please give a detailed description about the runtime environment? For example, what is the version of Python, Tensorflow and Matlab.
I tested the model on some images. I rotated the segment map as well. But, I can't get good results as shown in the paper. I have used the following repositories for pose model and human parser.
Pose Model : https://github.com/tensorboy/pytorch_Realtime_Multi-Person_Pose_Estimation
Human Parsing : https://github.com/Engineering-Course/LIP_JPPNet
I haven't gone through the code yet. Do I have to do anything more to get a good result? Has anyone got results the same as mentioned in the paper? Are there any restrictions on outputs of pose and human parsing models like fixed resolution, etc.,?
No such file or directory: 'data/viton_test_pairs.txt'
Hi,
Can you please share the annotated training data for FCN, and the FCN source code as well if possible.
Thanks
What is the format of the mat file needed for input images? can someone please specify?
A declarative, efficient, and flexible JavaScript library for building user interfaces.
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
An Open Source Machine Learning Framework for Everyone
The Web framework for perfectionists with deadlines.
A PHP framework for web artisans
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
Some thing interesting about web. New door for the world.
A server is a program made to process requests and deliver data to clients.
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
Some thing interesting about visualization, use data art
Some thing interesting about game, make everyone happy.
We are working to build community through open source technology. NB: members must have two-factor auth.
Open source projects and samples from Microsoft.
Google ❤️ Open Source for everyone.
Alibaba Open Source for everyone
Data-Driven Documents codes.
China tencent open source team.