ofa-sys / daflow Goto Github PK

May you please release the ckpt trained on Fashion Video dataset for testing? Besides, which is the difference between image animation and vitual try-on task on the network's input-output and structure?

The training result is blank。

Hi!
Kindly let me know about the required preprocessing steps to test on custom data.
And how to get key points for a person?

Hi,

Thank you for this amazing repository. I had one question about implementation of VGG19() in external_function.py, If you look at line 362-363, shouldn't it be "relu3_3":
for x in range(14, 16): self.relu3_3.add_module(str(x), features[x]) ,
currently it is written as self.relu3_2.add_module(str(x), features[x]) and has been repeated twice.

Thank You!

How to create custom dataset, to test for custom images

python -u test_SDAFNet_viton.py -b 8 --name TEST_PAIR --mode test --dataset_list VITON/test_pairs.txt --workers 0

May I ask if anyone can use the code and training data provided by the author to train a model that is consistent with the model testing results provided by the author? I used the code and training data provided by the author to obtain the model, but I am unable to obtain the testing results of the model provided by the author

cann you release your CLIP agin,thx

Thank you for sharing the excellent work! Is there any algorithm to generate the agnostic images or the images are cropped manually?

The image output from the training is fine, but using the retrained model for testing, the effect is very bad. What is the reason for this?

Before img_agnostic, how can I get the image under image-parse-new? Which human parse model should I use ?

Hi, thank you very much for your nice repository and for providing the relevant models and data. Can you please share how did you make the agnostic image inputs so that I can test this model on custom images? Thank you.

Hi! This work presents a nice Garment Warping Job！ However，when I test the fid on VITON under unpaired setting as CP-VTON+, I get the result 10.53, which is different from the result in your paper.

All my steps are in accordance with Readme.md, can you tell me where is the problem with my setting?

thx.

Dear @ShuaiBai623 ,

Thanks again for this work and share. In the paper, I see the following line "In particular, for a fair comparison, we train SDAFN+ model like PFAPN...". One advantage I see of using something like SDAFN+ is that we don't have to agnostic images or pose images as input during inferencing. (Apart from your research. goals of comparing with PFAFPN..) . Could you please let know if indeed you have a trained SDAFN+ model shared publicly?

Thanks,
Arun

The training process result graph is blank，can you help me?

Hello,is "img_agnostic" generated based on the image of the person or made by yourself?

Could you please provide a environment.yml for a conda environment? thank you in advance

Hi, thanks for the awesome work. I think are a few differences between the code and the paper.
Fig.3 in the paper shows that Self-MFE only takes in the reference features (or warped reference features). And the cross-MFE takes Source features and warped reference features as inputs. However, in the code, Self-MFE takes warped source features and warped reference features (not shown in Fig.3) as inputs as shown below:
input_feat = torch.cat([att_source_feat,att_reference_feat],1)

Also, the Cross-MFE takes in the raw reference features (instead of warped as shown in Fig.3) and warped source features of layer n-1.
input_feat = torch.cat([att_source_feat,feat_ref],1)

Am I getting this correctly? And is there any particular reason for these changes?

Used the provided ckpt_viton.pt model file and dataset from the author, but the test results were significantly different from what the author had published. Has anyone experienced a similar situation?In the author's article, they claimed that SSIM could reach 0.85 in the readme.md, but the results I obtained through testing were only around 0.6.

I want to run your code: data/data_process/get_img_agnostic.py

But I found that I didn't know what input it needed

Can you tell me, thank you!

Hello, I want to know why the pre-training model is replaced by the post-training model. There are results after the test, but the result chart is empty. I am a beginner. There may be some things that need to be changed or I don't quite understand what I am reading, i would appreciate it if you could tell me what the reason is

Hi, I hope that you are well.
When do we actually synthesise clothing on the individual when the reference person is not passed into the network

ref_input = torch.cat((pose, img_agnostic), dim=1)

 result_tryon, results_all = net(ref_input, cloth_img, img_agnostic, return_all=True)