Closed Wangxuefeng92 closed 9 months ago
Thank you for bringing up this issue. It appears there are some version incompatibilities within the released code. I will investigate this matter and release an updated version shortly.
thank you very much for your rely!!! I fixed these issues above and run this code. But unfortunatly,I got divergent result in ubc fashion dataset. Is there other problem in training?
I update the train.py
This version should work normally.
The former version uses fine-grained image prompts during training, but it is unstable.
We use null text prompts in the updated version, like MagicAnimate.
very appreciated for your help!thank you!
@Wangxuefeng92 Hi, I'd like to know do you use batch_size=1 or >1, I noticed that there are some comments in python file as "FIXME: only support train_batch_size=1".
hi,sir,thank your great project!I am trying your ’train.py‘, but there exist two issues: a. the 'dwpose_conditions' from dwpose is numpy-type, but model need is torch-type b. image_prompts is not used in training interface
Is there any miss in your uploaded code?