Dániel Varga

@RaananHadar I can compile the computation graph on 32GB RAM machines. The compile time for N=50 is just intolerable, 4 hours. I've seen it go above 50% RAM utilization, so...

@RaananHadar With current TH and TF it's a trade-off, really. With TF, you save lots and lots of compile time, and can fit into your 16GB core RAM. But I...

> Even if InceptionV1 is an old model, it can still be useful for some research work. Seconded. Here's some work we did with exactly this model, thanks @mdda! https://users.renyi.hu/~daniel/lucid/...

@kashif FYI it's not there anymore.

Got it, thank you! I've uploaded it to http://people.mokk.bme.hu/~daniel/vgg_face_model_weights.h5 @kashif Is this okay with you? If yes: I won't give any hard promises that the above link stays live for...

Sorry for the late reply! If it's still relevant: what's the extension of this image file? Are native libs like libpng properly installed?

Can you open the specific image where the process fails, like this?: ``` from scipy.misc import imread print imread("path/to/image.png").shape ```

Sorry for the late reply! Could you please post the contents of your .json file? Was it created with standard settings of train.py, together with the indentically named .h5?

That's weird, that model.json has the right number of layers. Could you maybe upload somewhere the corresponding h5 file as well?

That's weird. The three files that you gave me work for me. Keras 1.2.2, both backends, both image_dim_orderings, actually all four combinations of these. `webcam.py` successfully starts with these files,...