MenghaoGuo / AutoDeeplab

Pytorch Implementation the paper Auto-DeepLab Hierarchical Neural Architecture Search for Semantic Image Segmentation

https://arxiv.org/abs/1901.02985

MenghaoGuo/AutoDeeplab Issues

AttributeError: 'Cell' object has no attribute 'cell_ops'
Updated a year ago2
TypeError: 'NoneType' object is not callable
Updated 2 years ago9
ModuleNotFoundError: No module named 'modeling'
Closed 4 years ago
ModuleNotFoundError
Updated 4 years ago2
How can I visualize the best inner cell and outer network after search for my custom dataset?
Updated 4 years ago
What does the index of alphas_network mean in auto_deeplab.py?
Updated 4 years ago1
Given groups=1, weight of size 48 160 1 1, expected input[2, 1280, 25, 25] to have 160 channels, but got 1280 channels instead
Updated 4 years ago
Still 'CUDA out of memory' with 4 TITAN X (pascal) when training model in PASCAL VOC dataset
Updated 4 years ago
About the model size
Updated 4 years ago3
how to obtain and train the target architecture of searc？
Updated 5 years ago2
memory consumption too large
Updated 5 years ago14
what does ABN in "from operations import ABN" in aspp.py ?
Closed 5 years ago
Not doing an architecture step until epoch > 19
Updated 5 years ago
train_autodeeplab.py: error: unrecognized arguments: --gpu_ids 0,1,2,3
Updated 5 years ago2
missing LICENSE file
Updated 5 years ago
Pytorch 1
Closed 5 years ago1
Very low mIOU value
Updated 5 years ago12
how many epoches should we set to make the train loss to be convergent?
Closed 5 years ago
about the crop image size for input
Closed 5 years ago7
self.train_loader1, self.train_loader2 in train_autodeeplab.py
Closed 5 years ago1
File "../AutoDeeplab-master/operations.py", line 90, in forward out = torch.cat([self.conv_1(x), self.conv_2(x[:,:,1:,1:])], dim=1)
Closed 5 years ago
File "../AutoDeeplab-master2/model_search.py", line 68, in forward s = sum(self._ops[offset+j](h, weights[offset+j]) for j, h in enumerate(states)) RuntimeError: The size of tensor a (81) must match the size of tensor b (82) at non-singleton dime
Closed 5 years ago
I think the decode part is just a single DFS rather than viterbi algorithm ∩_∩, and definitely can be optimized.
Updated 5 years ago
why we should multiply softmax result by 5
Closed 5 years ago4
Use GPU '0,1,2,3', but only the card 0 is actually used.
Updated 5 years ago7
Network Architecture Weights are non zero for connections that do not exist
Updated 5 years ago
The network weights and the architecture weights(coefficient) train together?
Updated 5 years ago2
performance report
Updated 5 years ago2
what is train_loader1 and train_loader2?
Updated 5 years ago6
Softmax applied twice to the archtecture parameters?
Closed 5 years ago1
arch_lr value not provided
Updated 5 years ago2
AttributeError
Updated 5 years ago5
Statement
Closed 5 years ago
Single GPU with workers=1, but I still met cuda runtime error.
Closed 5 years ago1
Where is utils？
Closed 5 years ago2
ValueError: not enough values to unpack (expected 5, got 4)
Closed 5 years ago2
how to do inference?
Updated 5 years ago
RuntimeError
Updated 5 years ago
no module named 'mypath'
Closed 6 years ago6
test
Updated 6 years ago
Are these total codes?
Closed 6 years ago1
How to implement data input？
Closed 6 years ago5