About Diffuser unet_2d_blocks.py

[hmxiong]

Great work！！ I have a question about code. When I use COCO Instance weights to generate images, the error "Tensors must have same number of dimensions: got 3 and 4" will be reported in model/diffusers/models/unet_2d_blocks.py. I am using the latest version of diffusers code and stable diffusion-1.4 , torch-1.10 + cu111 . The following is my error message:

File "/home/xhm/Desktop/Code/Diffusion/DatasetDM/parallel_generate_Instance_COCO_AnyClass.py", line 412, in sub_processor images_here, x_t = ptp_utils.text2image(unet,vae,tokenizer,text_encoder,scheduler, prompts, controller, num_inference_steps=NUM_DIFFUSION_STEPS, guidance_scale=5, generator=g_cpu, low_resource=LOW_RESOURCE, Train=False) File "/home/xhm/anaconda3/envs/datasetdm/lib/python3.8/site-packages/torch/autograd/grad_mode.py", line 28, in decorate_context return func(*args, kwargs) File "/home/xhm/Desktop/Code/Diffusion/DatasetDM/ptp_utils.py", line 212, in text2image latents = diffusion_step_DDM(Unet, scheduler, controller, latents, context, t, guidance_scale, low_resource) File "/home/xhm/Desktop/Code/Diffusion/DatasetDM/ptp_utils.py", line 141, in diffusion_step_DDM noise_pred = unet(latents_input, t, encoder_hidden_states=context)["sample"] File "/home/xhm/anaconda3/envs/datasetdm/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1102, in _call_impl return forward_call(*input, *kwargs) File "/home/xhm/Desktop/Code/Diffusion/DatasetDM/model/unet.py", line 144, in forward sample, up_samples = upsample_block( File "/home/xhm/anaconda3/envs/datasetdm/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1102, in _call_impl return forward_call(input, kwargs) File "/home/xhm/Desktop/Code/Diffusion/DatasetDM/model/diffusers/models/unet_2d_blocks.py", line 2202, in forward hidden_states = torch.cat([hidden_states, res_hidden_states], dim=1) RuntimeError: Tensors must have same number of dimensions: got 3 and 4

I'm really sorry, I can't take a screenshot due to my computer.Thank you so much！！！

[hkxiao]

I meet the same question. @hmxiong

[weijiawu]

I suspect that the error is caused by the version of Diffuser. Perhaps you can try replacing the files under ./model with Diffuser version 0.3.0 （https://pypi.org/project/diffusers/0.3.0/#files） and see if that resolves the issue.

Thanks for your attention~

[hmxiong]

Sorry, this method did not solve my problem. Here I will show the code analysis as follows： in /diffusers/models/unet_blocks.py UpBlock2D Function only reture the hidden_states ，but in /model/unet.py you need both sample and up_samples. I think by solving this problem, I can solve mine.

[weijiawu]

We have already updated the diffuser at https://github.com/showlab/DatasetDM/tree/main/model/diffusers. You can directly use this version. Hope this can be helpful to you.

[hmxiong]

Thanks for sharing, the problem has been solved！！