Training MVTec Wood datasets & Different inference results about dfmgan HOT 3 CLOSED

jmlee99o commented on August 29, 2024

Training MVTec Wood datasets & Different inference results

from dfmgan.

Comments (3)

Ldhlwh commented on August 29, 2024

What parameters did you handle? (It may be a little rude, but I'm most curious about it.)

If you mean hyperparameters, we have presented all our choices in the codes and the paper. If you mean model parameters, you may have a look at the codes below.

DFMGAN/training/training_loop.py

Lines 234 to 290 in abe6d64

 for name, module, opt_kwargs, reg_interval in training_nets: 

 num_param = 0 

 if name == 'D': 

 if G_kwargs.transfer == 'res_block_uni_dis': 

 for res in [4, 8, 16, 32, 64, 128, 256]: 

 target_param = [] 

 if res > D_kwargs.uni_st: 

 num_param += sum([p.numel() for p in getattr(module, f'mask_b{res}').parameters()]) 

 target_param.append(getattr(module, f'mask_b{res}').parameters()) 

 else: 

 num_param += sum([p.numel() for p in getattr(module, f'uni_b{res}').parameters()]) 

 target_param.append(getattr(module, f'uni_b{res}').parameters()) 

 target_param = itertools.chain(*target_param) 

 else: 

 num_param = sum([p.numel() for p in module.parameters()]) 

 target_param = module.parameters() 

 elif name == 'D_match': 

 num_param = sum([p.numel() for p in module.parameters()]) 

 target_param = module.parameters() 

 elif name == 'G': 

 if ft == 'default': 

 num_param = sum([p.numel() for p in module.parameters()]) 

 target_param = module.parameters() 

 elif ft == 'ft_map': 

 num_param = sum([p.numel() for p in module.mapping.parameters()]) 

 target_param = module.mapping.parameters() 

 elif ft == 'ft_syn': 

 num_param = sum([p.numel() for p in module.synthesis.parameters()]) 

 target_param = module.synthesis.parameters() 

 elif ft.startswith('ft_syn_'): 

 num_trainable_block = int(ft.split('_')[-1]) 

 syn_modules = [module.synthesis.b4, module.synthesis.b8, module.synthesis.b16, module.synthesis.b32, module.synthesis.b64, module.synthesis.b128, module.synthesis.b256] 

 target_param = itertools.chain(*[mod.parameters() for mod in syn_modules[:num_trainable_block]]) 

 num_param = sum([p.numel() for p in target_param]) 

 target_param = itertools.chain(*[mod.parameters() for mod in syn_modules[:num_trainable_block]]) 

 elif ft.startswith('ft_map_syn_'): 

 num_trainable_block = int(ft.split('_')[-1]) 

 syn_modules = [module.synthesis.b4, module.synthesis.b8, module.synthesis.b16, module.synthesis.b32, module.synthesis.b64, module.synthesis.b128, module.synthesis.b256] 

 target_param = itertools.chain(*[mod.parameters() for mod in syn_modules[:num_trainable_block]], module.mapping.parameters()) 

 num_param = sum([p.numel() for p in target_param]) 

 target_param = itertools.chain(*[mod.parameters() for mod in syn_modules[:num_trainable_block]], module.mapping.parameters()) 

 elif ft == 'transfer': 

 if G_kwargs.transfer == 'dual_mod': 

 target_param = module.defect_mapping.parameters() 

 num_param = sum([p.numel() for p in target_param]) 

 target_param = module.defect_mapping.parameters() 

 elif G_kwargs.transfer in ['res_block', 'res_block_match_dis', 'res_block_uni_dis']: 

 target_param = [module.defect_mapping.parameters()] 

 num_param += sum([p.numel() for p in module.defect_mapping.parameters()]) 

 for res in [4, 8, 16, 32, 64, 128, 256]: 

 if res >= G_kwargs.synthesis_kwargs.res_st: 

 target_param.append(getattr(module.synthesis, f'res_b{res}').parameters()) 

 num_param += sum([p.numel() for p in getattr(module.synthesis, f'res_b{res}').parameters()]) 

 target_param = itertools.chain(*target_param)

2. Did you take other Augmentation tech except for ADA of StyleGAN2 - ADA ?

No, just ADA.

3. This is different from the point, when I generate a defective image in my dataset, in that model, both normal and defect images are generated and combined from Latent vector, but can I modify the model by inputting it into the normal image I have without generating the generated image and combining it with the generated defective image? (just generating defect images from latent vectors, but not for normal images. because, I think generated normal images of my industrial dataset will have poor quaility.)

If you believe you are not able to get a good base model generating defect-free images in the first stage, then our DFMGAN may not meet your requirement since it depends on well-learned object/texture features. You may consider anomaly generation methods of image-to-image pipeline like Defect-GAN.

from dfmgan.

lu1zero9 commented on August 29, 2024

Hi, I am graduate student in korea. I want to generate various defective industrial data images by utilizing your model, so before handling our datas, I use MVTec Wood datasets (Defect : Scratch) to confirm and check the model's performances.

But, when I train model and inference with wood datasets, Its result is quite different with your paper's result. Both Good and Mask images have not good quality than I expected. (Furthermore, It occurs Overfitting after 400 kimg in good images / Generated Defect images are not similar with MVTec dataset's original scratch things)

I want to know how to optimize the parameters ( that you used ) to develop image qualities.

What parameters did you handle? (It may be a little rude, but I'm most curious about it.)

Did you take other Augmentation tech except for ADA of StyleGAN2 - ADA ?

This is different from the point, when I generate a defective image in my dataset, in that model, both normal and defect images are generated and combined from Latent vector, but can I modify the model by inputting it into the normal image I have without generating the generated image and combining it with the generated defective image? (just generating defect images from latent vectors, but not for normal images. because, I think generated normal images of my industrial dataset will have poor quaility.)

Thanks for reading! I appreciate for your model services.

Hi, have you generated all the results to check the effects? If so, could you please send me a copy of the generated results? I would like to take a look. Thank you very much, I appreciate it!

from dfmgan.

Ldhlwh commented on August 29, 2024

I'll close this issue for now since it's been inactive for weeks.

from dfmgan.

Training MVTec Wood datasets & Different inference results about dfmgan HOT 3 CLOSED

Comments (3)

Related Issues (20)

Recommend Projects

React

Vue.js

Typescript

TensorFlow

Django

Laravel

D3

Recommend Topics

javascript

web

server

Machine learning

Visualization

Game

Recommend Org

Facebook

Microsoft

Google

Alibaba

D3

Tencent

	for name, module, opt_kwargs, reg_interval in training_nets:

	num_param = 0
	if name == 'D':
	if G_kwargs.transfer == 'res_block_uni_dis':
	for res in [4, 8, 16, 32, 64, 128, 256]:
	target_param = []
	if res > D_kwargs.uni_st:
	num_param += sum([p.numel() for p in getattr(module, f'mask_b{res}').parameters()])
	target_param.append(getattr(module, f'mask_b{res}').parameters())
	else:
	num_param += sum([p.numel() for p in getattr(module, f'uni_b{res}').parameters()])
	target_param.append(getattr(module, f'uni_b{res}').parameters())
	target_param = itertools.chain(*target_param)
	else:
	num_param = sum([p.numel() for p in module.parameters()])
	target_param = module.parameters()

	elif name == 'D_match':
	num_param = sum([p.numel() for p in module.parameters()])
	target_param = module.parameters()

	elif name == 'G':
	if ft == 'default':
	num_param = sum([p.numel() for p in module.parameters()])
	target_param = module.parameters()
	elif ft == 'ft_map':
	num_param = sum([p.numel() for p in module.mapping.parameters()])
	target_param = module.mapping.parameters()
	elif ft == 'ft_syn':
	num_param = sum([p.numel() for p in module.synthesis.parameters()])
	target_param = module.synthesis.parameters()
	elif ft.startswith('ft_syn_'):
	num_trainable_block = int(ft.split('_')[-1])
	syn_modules = [module.synthesis.b4, module.synthesis.b8, module.synthesis.b16, module.synthesis.b32, module.synthesis.b64, module.synthesis.b128, module.synthesis.b256]
	target_param = itertools.chain(*[mod.parameters() for mod in syn_modules[:num_trainable_block]])
	num_param = sum([p.numel() for p in target_param])
	target_param = itertools.chain(*[mod.parameters() for mod in syn_modules[:num_trainable_block]])
	elif ft.startswith('ft_map_syn_'):
	num_trainable_block = int(ft.split('_')[-1])
	syn_modules = [module.synthesis.b4, module.synthesis.b8, module.synthesis.b16, module.synthesis.b32, module.synthesis.b64, module.synthesis.b128, module.synthesis.b256]
	target_param = itertools.chain(*[mod.parameters() for mod in syn_modules[:num_trainable_block]], module.mapping.parameters())
	num_param = sum([p.numel() for p in target_param])
	target_param = itertools.chain(*[mod.parameters() for mod in syn_modules[:num_trainable_block]], module.mapping.parameters())
	elif ft == 'transfer':
	if G_kwargs.transfer == 'dual_mod':
	target_param = module.defect_mapping.parameters()
	num_param = sum([p.numel() for p in target_param])
	target_param = module.defect_mapping.parameters()
	elif G_kwargs.transfer in ['res_block', 'res_block_match_dis', 'res_block_uni_dis']:
	target_param = [module.defect_mapping.parameters()]
	num_param += sum([p.numel() for p in module.defect_mapping.parameters()])
	for res in [4, 8, 16, 32, 64, 128, 256]:
	if res >= G_kwargs.synthesis_kwargs.res_st:
	target_param.append(getattr(module.synthesis, f'res_b{res}').parameters())
	num_param += sum([p.numel() for p in getattr(module.synthesis, f'res_b{res}').parameters()])
	target_param = itertools.chain(*target_param)