I use DeepFashion Dataset，then run python train.py --name=DPTN_fashion --model=DPTN --dataset_mode=fashion --dataroot=./dataset/fashion --batchSize 32 --gpu_id=0 At first the loss was falling and then rising again This is my train_opt.txt: affine: True batchSize: 8 beta1: 0.5 checkpoints_dir: ./checkpoints continue_train: False data_type: 32 dataroot: ./dataset/fashion dataset_mode: fashion debug: False device: cuda dis_layers: 4 display_env: DPTNfashion display_freq: 200 display_id: 0 display_port: 8096 display_single_pane_ncols: 0 display_winsize: 512 feat_num: 3 fineSize: 512 fp16: False gan_mode: lsgan gpu_ids: [0] image_nc: 3 init_type: orthogonal input_nc: 3 instance_feat: False isTrain: True iter_start: 0 label_feat: False label_nc: 35 lambda_content: 0.25 lambda_feat: 10.0 lambda_g: 2.0 lambda_rec: 2.5 lambda_style: 250 layers_g: 3 loadSize: 256 load_features: False load_pretrain: load_size: 256 local_rank: 0 lr: 0.0002 lr_policy: lambda max_dataset_size: inf model: DPTN nThreads: 2 n_clusters: 10 n_downsample_E: 4 n_layers_D: 3 name: DPTN_fashion ndf: 64 nef: 16 nhead: 2 niter: 100 niter_decay: 100 no_flip: False no_ganFeat_loss: False no_html: False no_instance: False no_vgg_loss: False norm: instance num_CABs: 2 num_D: 1 num_TTBs: 2 num_blocks: 3 old_size: (256, 176) output_nc: 3 phase: train pool_size: 0 pose_nc: 18 print_freq: 200 ratio_g2d: 0.1 resize_or_crop: scale_width save_epoch_freq: 1 save_input: False save_latest_freq: 1000 serial_batches: False structure_nc: 18 t_s_ratio: 0.5 tf_log: False use_coord: False use_dropout: False use_spect_d: True use_spect_g: False verbose: False which_epoch: latest

Do I need to continue training or do I stop to change the parameters. Thank You！

Hello, I am wondering how to test on unseen image. I followed the procedure and I found I need to generate keypoints for my images. I found the PATN project >> tool/compute_coordiantes.py. But when I run this file on my image, there is an error says the size mismatch which caused at line 216. (output1, output2 = model.predict(imageToTest_padded)). It looks like because of the multiplier, the input sizes are different, so there always an error. I am wondering how to solve this problem?

Thank you!

Hi, can you check if the data set I prepared is correct? I cannot understand GFLA's evaluation setting.

I did, [gt_path] dataset : (750 x 1101) -> (176, 256). *using PIL.Image.resize((176, 256)) [fid_real_path] dataset : (750 x 1101) -> (176, 256). *using PIL.Image.resize((176, 256))

Every scores are lower than on paper. Do I have to resize images by usingcv2.resize method?

Hello, I am interested in learning about pose translation from one image to another ,and I came across your repository. I would like to know

In readme it's mentioned

Download the DeepFashion dataset [in-shop clothes retrival benchmark](http://mmlab.ie.cuhk.edu.hk/projects/DeepFashion/InShopRetrieval.html), and put them under the ./dataset/fashion directory.

Download train/test pairs and train/test keypoints annotations from [Google Drive] including fasion-resize-pairs-train.csv, fasion-resize-pairs-test.csv, fasion-resize-annotation-train.csv, fasion-resize-annotation-train.csv, train.lst, test.lst, and put them under the ./dataset/fashion directory.

Split the raw image into the training set (./dataset/fashion/train) and test set (./dataset/fashion/test)

As such ,in order to try the repository on custom dataset are these the steps which should be taken.

1. Split into train test ,crop and Then execute

python tool/generate_fashion_datasets.py

2. Use openpose to obtain keypoints.apart from openpose can we use any other keypoint estimation library. Is it mandatory to use openpose to obtain keypoints.Can any other pose estimation framework like mediapipe be used
3. ,then create pairs.csv using

python2 tool/create_pairs_dataset.py

Apart from these steps,am I missing anything

• Your work inspires me a lot！
• It is mentioned in the text that train the source-to-source network as an auxiliary, and then share the weights to the source-to-target, I would like to know their training steps?
• First ignore the Pose Transformer Module, train the source-to-source network, then share parameters to the source-to-target network, and finally train the Pose Transformer Module.
• Is that what I think? I want to know more details！
• Thank u a lot!

Hey @PangzeCheung ,

Again me, hope everything goes well with you.

1. I noticed the niter and niter_decay are all set to 100. That's to say we need to train the entire model totally 200 epochs for each dataset?
2. Since I also find that the pre-trained checkpoints you provided for DeepFashion is at 190 epoch, while for market-1501, it is just iter 811200. So I am a bit confused on how should I choose a checkpoint between checkpoint stored by epoch or stored by total iters?
3. I alreadt succefully trained the Dual-Task PTN with 1 Nvidia TITAN xp gpu on Market-1501 dataset, it cost me around 4 days by now (Stil at Epoch 131). But there all lots of checkpoints named in two different ways as follows. I also want to ask is this all right?

Looking forward to you reply but not in a hurry~

Best Regards,

Hey @PangzeCheung , Thanks for your impressive work, It's really interesting, I want to reproduce the results. However, when I want to re-train the model by meself, I find there should one files named base_option.py missed. Could you kindly add the file?

Many Thanks

First of all, thanks for making the code public. I have prepared the DeepFashion dataset following the instructions in the README, downloaded the trained model, and tried the test code, but I did not get the results as published in the paper. is the README correct? Also, from the script for preparing the dataset, it seems that it is trained on low resolution images, will the code and trained models for the high resolution DeepFashion dataset be published?

How many epochs should I train to get the results you shown in the paper. It seems need 140 epochs, which will cost 1 week. I use one Tesla V100-32g in training time, and when using multiple gpus, there will be serious load imbalance.