neuralhash
neuralhash copied to clipboard
What are the steps to run this project
Can you please update on how to run this project?
Can you please update on how to run this project?
hello, how to run this project?
While I don't have this project with respect to the correct results. I will document what got me starting. The code suggests that colornet was used as input data to train the model. This dataset (1.8TB full size) can be obtained via http://places.csail.mit.edu/downloadData.html Instead I used 1000 photos from my own collection then scaled them down due to memory constraints.
mkdir -p jobs
mkdir -p output
mkdir -p data/colornet
mogrify -resize 256x256\! -path /your/path/to/neuralhash/data/colornet /your/path/to/photos/*jpg
At this moment I have removed all references to the VisdomLogger, but I would expect running it and changing the IP adress would obviously help too. Because my only Cuda machine had only 4GB of RAM I changed utils.py BATCH_SIZE=2. I also commented out the model2 testing while training (because I ran out of memory).
So once you have done the steps above you are able to run train_amnesia.py. This would give the weights_file mentioned in the rest of the code.
Where I am currently at:
python encoding.py /tmp/test-small.jpg --model=output/train_test.pth --max_iter=500 /tmp/test2.png
Target: 11100001101111111101100001100011
Changing distribution size: 96 -> 96
/home/skinkie/.local/lib/python3.9/site-packages/torch/nn/functional.py:3487: UserWarning: nn.functional.upsample is deprecated. Use nn.functional.interpolate instead.
warnings.warn("nn.functional.upsample is deprecated. Use nn.functional.interpolate instead.")
/home/skinkie/.local/lib/python3.9/site-packages/torch/nn/functional.py:718: UserWarning: Named tensors and all their associated APIs are an experimental feature and subject to change. Please do not use them for anything important until they are released as stable. (Triggered internally at /pytorch/c10/core/TensorImpl.h:1156.)
return torch.max_pool2d(input, kernel_size, stride, padding, dilation, ceil_mode)
Fixing distribution size: 96 -> 96
skinkie@aspire7:~/neuralhash$ cat decoding.py
import sys
from utils import *
from models import DecodingModel, DataParallelModel
import transforms
model = DataParallelModel(DecodingModel.load(distribution=transforms.encoding, n=96, weights_file="output/train_test.pth"))
image = im.torch(im.load('/tmp/test2.png')).unsqueeze(0)
# target = binary.parse(str(target))
prediction = model(image).mean(dim=1).squeeze().cpu().data.numpy()
prediction = binary.get(prediction)
print (f"Prediction: {binary.str(prediction)}")
skinkie@aspire7:~/neuralhash$ python decoding.py
/home/skinkie/.local/lib/python3.9/site-packages/torch/nn/functional.py:3487: UserWarning: nn.functional.upsample is deprecated. Use nn.functional.interpolate instead.
warnings.warn("nn.functional.upsample is deprecated. Use nn.functional.interpolate instead.")
/home/skinkie/.local/lib/python3.9/site-packages/torch/nn/functional.py:718: UserWarning: Named tensors and all their associated APIs are an experimental feature and subject to change. Please do not use them for anything important until they are released as stable. (Triggered internally at /pytorch/c10/core/TensorImpl.h:1156.)
return torch.max_pool2d(input, kernel_size, stride, padding, dilation, ceil_mode)
Prediction: 11110001101011101000001111101011
And as you can see, the Target != Prediction.
Now I have noticed that in models.py the following is set. Where MODEL_TYPE in utils.py is defined as DecodingGramNet.
DecodingModel = eval(MODEL_TYPE)
Even changing this model to DecodingNet does not result in a correct prediction. And TinyDecodingNet failing when encoding. And DilatedDecodingNet failing on usage (due to GramMatrix not being defined).
So from what I have experienced this thing does not work out of the box, not even when investing some time in it.