fast-style-transfer
fast-style-transfer copied to clipboard
lengstrom
Can't get style.py to do anything!
I'm a total noob - so apologies in advance, I've hopefully installed everything correctly - but when I run style.py there is no output - and the process takes a very suspiciously short time.
Here's the call python style.py --checkpoint-dir C:\Users\ga_ma\Desktop\DeepArt\fast-style-transfer-master --style C:\Users\ga_ma\Desktop\DeepArt\fast-style-transfer-master\examples\style\wave.jpg --test C:\Users\ga_ma\Desktop\DeepArt\fast-style-transfer-master\examples\content\stata.jpg --test-dir C:\Users\ga_ma\Desktop\DeepArt\fast-style-transfer-master --content-weight 1.5e1 --checkpoint-iterations 1000 --batch-size 20 --vgg-path C:\Users\ga_ma\Desktop\DeepArt\fast-style-transfer-master\data\imagenet-vgg-verydeep-19.mat
and here's the output from the terminal
(base) C:\Users\ga_ma\Desktop\DeepArt\fast-style-transfer-master>python style.py --checkpoint-dir C:\Users\ga_ma\Desktop\DeepArt\fast-style-transfer-master --style C:\Users\ga_ma\Desktop\DeepArt\fast-style-transfer-master\examples\style\wave.jpg --test C:\Users\ga_ma\Desktop\DeepArt\fast-style-transfer-master\examples\content\stata.jpg --test-dir C:\Users\ga_ma\Desktop\DeepArt\fast-style-transfer-master --content-weight 1.5e1 --checkpoint-iterations 1000 --batch-size 20 --vgg-path C:\Users\ga_ma\Desktop\DeepArt\fast-style-transfer-master\data\imagenet-vgg-verydeep-19.mat
(1, 514, 928, 3)
2018-07-20 16:59:43.855680: I T:\src\github\tensorflow\tensorflow\core\platform\cpu_feature_guard.cc:141] Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX2
2018-07-20 16:59:44.862621: I T:\src\github\tensorflow\tensorflow\core\common_runtime\gpu\gpu_device.cc:1392] Found device 0 with properties:
name: GeForce GTX 1050 major: 6 minor: 1 memoryClockRate(GHz): 1.493
pciBusID: 0000:01:00.0
totalMemory: 4.00GiB freeMemory: 3.30GiB
2018-07-20 16:59:44.870256: I T:\src\github\tensorflow\tensorflow\core\common_runtime\gpu\gpu_device.cc:1471] Adding visible gpu devices: 0
2018-07-20 16:59:45.518940: I T:\src\github\tensorflow\tensorflow\core\common_runtime\gpu\gpu_device.cc:952] Device interconnect StreamExecutor with strength 1 edge matrix:
2018-07-20 16:59:45.522176: I T:\src\github\tensorflow\tensorflow\core\common_runtime\gpu\gpu_device.cc:958] 0
2018-07-20 16:59:45.524477: I T:\src\github\tensorflow\tensorflow\core\common_runtime\gpu\gpu_device.cc:971] 0: N
2018-07-20 16:59:45.526639: I T:\src\github\tensorflow\tensorflow\core\common_runtime\gpu\gpu_device.cc:1084] Created TensorFlow device (/job:localhost/replica:0/task:0/device:GPU:0 with 3025 MB memory) -> physical GPU (device: 0, name: GeForce GTX 1050, pci bus id: 0000:01:00.0, compute capability: 6.1)
2018-07-20 16:59:55.793411: I T:\src\github\tensorflow\tensorflow\core\common_runtime\gpu\gpu_device.cc:1471] Adding visible gpu devices: 0
2018-07-20 16:59:55.795700: I T:\src\github\tensorflow\tensorflow\core\common_runtime\gpu\gpu_device.cc:952] Device interconnect StreamExecutor with strength 1 edge matrix:
2018-07-20 16:59:55.798504: I T:\src\github\tensorflow\tensorflow\core\common_runtime\gpu\gpu_device.cc:958] 0
2018-07-20 16:59:55.800505: I T:\src\github\tensorflow\tensorflow\core\common_runtime\gpu\gpu_device.cc:971] 0: N
2018-07-20 16:59:55.803158: I T:\src\github\tensorflow\tensorflow\core\common_runtime\gpu\gpu_device.cc:1084] Created TensorFlow device (/job:localhost/replica:0/task:0/device:GPU:0 with 3025 MB memory) -> physical GPU (device: 0, name: GeForce GTX 1050, pci bus id: 0000:01:00.0, compute capability: 6.1)
UID: 33
Training complete. For evaluation:
python evaluate.py --checkpoint C:\Users\ga_ma\Desktop\DeepArt\fast-style-transfer-master ...
I dont know how long you've trained model ,but it looks like success to train your model.
Checkout your ./checkpoints dir and you can find what you want
It doesn't look like you've downloaded the COCO train2014 folder, which contains the image files needed to train the model. Use "setup.sh" to download and unzip that.
Use the argument --train-path to change the path to wherever your unzipped train2014 folder is.
I would also recommend changing your --batch-size from 20 to the default 4, since having such a high bach size will result in the extreme trimming of the dataset, and may also cause this "extremely fast training" issue.
I do know that python generally dislikes backslashes, try and use / instead
I am working on a L-BFGS optimizer update for this repo, that may fix your issue since this is using the ADAM optimizer.
just wanted to comment that I was having the same issue even though I had the coco 2014 data. My issue was I was pointing to the train2014.zip when I needed to point to the unzipped file. hope this is helpful to someone ^-^
