The new pretrained model released on 9th Sep caused corrupt when load

[huochaitiantang]

The new pretrained model released on 9th Sep caused corrupt when load

[exition]

me too , wrong encoded pth file

[xiaoqiangsheng2016]

UnicodeDecodeError: 'utf-8' codec can't decode byte 0xda in position 5: invalid continuation byte

me too

[huochaitiantang]

Hey guys, the bug is fixed now, and the patch is upstreamed!
Please see:

pytorch-deep-image-matting/core/demo.py

Lines 10 to 31 in f5f909d

	'''
	Load model exported by python2 with python3 will cause error:
	ascii' codec can't decode byte 0xda in position 5
	The following codes will fix the bug
	'''
	def my_torch_load(fname):
	try:
	ckpt = torch.load(fname)
	return ckpt
	except Exception as e:
	print("Load Error:{}\nTry Load Again...".format(e))
	class C:
	pass
	def c_load(ss):
	return pickle.load(ss, encoding='latin1')
	def c_unpickler(ss):
	return pickle.Unpickler(ss, encoding='latin1')
	c = C
	c.load = c_load
	c.Unpickler = c_unpickler
	ckpt = torch.load(args.resume, pickle_module=c)
	return ckpt