ProGamerGov

Has anyone been able to match Dreamscope results?

26

I can't seem to get close. If you have, what sort of settings are you running? I shamelessly copied the question from https://github.com/jcjohnson/neural-style/issues/240

``` th caffemodel_to_t7.lua -output_t7 VGG_ILSVRC_19_layers.t7 -input_caffemodel VGG_ILSVRC_19_layers.caffemodel -input_prototxt VGG_ILSVRC_19_layers_deploy.prototxt python t7_to_state_dict.py --input_t7 VGG_ILSVRC_19_layers.t7 --model_name vgg19 python test_model.py --t7_file VGG_ILSVRC_19_layers.t7 --pth_file vgg19-bdbb0f7f.pth ``` Upon running the test_model.py script, I get this...

cudnnConvolutionBackwardData failed - Error in CuDNN: CUDNN_STATUS_NOT_SUPPORTED (cudnnConvolutionBackwardData)

8

I'm not sure what is causing this error, and how to fix it: ``` cudnnConvolutionBackwardData failed: 9 convDesc=[mode : CUDNN_CROSS_CORRELATION datatype : CUDNN_DATA_FLOAT] hash=-dimA1,3,2615,2816 -filtA64,3,3,3 1,64,2615,2816 -padA1,1 -convStrideA1,1 CUDNN_DATA_FLOAT /home/ubuntu/torch/install/bin/luajit:...

ValueError: Incorrect number of features. Got 4 features, expected 3

4

``` ubuntu@ip-Address:~/fast-neural-doodle$ python get_mask_hdf5.py --n_colors=5 --style_image=Monet2.jpg --style_mask=Monet2_sem2_r.png --target_mask=MonetNew2_sem2.png Traceback (most recent call last): File "get_mask_hdf5.py", line 45, in labels_target = kmeans.predict(target_flatten.astype(float)) File "/usr/lib/python2.7/dist-packages/sklearn/cluster/k_means_.py", line 889, in predict X = self._check_test_data(X)...

Add Captum to Pytorch & related libraries

1

Captum is one of the official PyTorch libraries, so I placed it under the link for the main PyTorch library. Let me know if I need to change anything!

Add missing None type hint to init functions

6

All `__init__` functions should have the type hint `-> None` according to PEP-0484: https://www.python.org/dev/peps/pep-0484/

- This should make model architecture detection more robust.

Bug in CLIPTokenizer input handling

6

## 🐛 Bug **Describe the bug** The output of OpenAI's CLIP tokenizer is different than Torchtext's tokenizer, when using the same inputs & settings. **To Reproduce** Steps to reproduce the...

`CLIPTokenizer` should output tensors in it's `forward` function rather than lists of numbers in str form

6

## 🚀 Feature Outputs for the current CLIP tokenizer appear to be a list of strings of numbers, rather than a tensor or even a list of numbers: ``` clip_tokenizer...

## 🐛 Bug It seems to me that the first line is ignored due to it being a comment about the file in open_clip and OpenAI's vocab file, but this...