generate unnorma images when inference

[Blackkinggg]

I can generated right images using the demo.py, but got below images when used the inference.py

[Blackkinggg]

There are also problems in the training stage.
Neural-Texture-Extraction-Distribution/util/visualization/linear_attention.py:

def attn2image(source_softmax, seg_query, input_image):
target_list = []
image_size = input_image.shape[2:]
for softmax, query in zip(source_softmax, seg_query):
b, num_label, h, w = query.shape
input_resize = F.interpolate( input_image, (h,w) )
input_resize = input_resize.view(b, -1, h*w)
extracted = torch.einsum('bkm,bvm->bvk', softmax, input_resize)
query = F.softmax(query.view(b, num_label, -1), 1)
estimated_target = torch.einsum('bkm,bvk->bvm', query, extracted)
estimated_target = estimated_target.view(b, -1, h, w)
target_list.append(F.interpolate(estimated_target, image_size))
target_gen = torch.cat(target_list, 3)
return target_gen

The source_softmax is an empty list.

[RenYurui]

Hi!
Is the checkpoint loaded successfully while inference?
Can you provide the messages when you run inference.py

[Blackkinggg]

Hi! Is the checkpoint loaded successfully while inference? Can you provide the messages when you run inference.py

您好，我使用的原始代码，看起来inference没有成功load模型，没有报错信息，直接生成上面的图像

[RenYurui]

When you run inference.py without training the model. You may need to load the model by adding --which_iter such as:

python -m torch.distributed.launch \
--nproc_per_node=1 \
--master_port 12345 inference.py \
--config ./config/fashion_512.yaml \
--name fashion_512 \
--no_resume \
--output_dir ./result/fashion_512/inference \
--which_iter 495400

[SHULJJ]

hello ，请问你解决了这个问题吗

[SHULJJ]

There are also problems in the training stage. Neural-Texture-Extraction-Distribution/util/visualization/linear_attention.py:

def attn2image(source_softmax, seg_query, input_image): target_list = [] image_size = input_image.shape[2:] for softmax, query in zip(source_softmax, seg_query): b, num_label, h, w = query.shape input_resize = F.interpolate( input_image, (h,w) ) input_resize = input_resize.view(b, -1, h*w) extracted = torch.einsum('bkm,bvm->bvk', softmax, input_resize) query = F.softmax(query.view(b, num_label, -1), 1) estimated_target = torch.einsum('bkm,bvk->bvm', query, extracted) estimated_target = estimated_target.view(b, -1, h, w) target_list.append(F.interpolate(estimated_target, image_size)) target_gen = torch.cat(target_list, 3) return target_gen

The source_softmax is an empty list.
请问你解决这个问题了吗

[shrinivasait]

Hi @Blackkinggg, How did generate pose key points?