Eyal Gruss
Eyal Gruss
in rnn, shouldn't: self.delta_W_rec = theano.shared(value = numpy.zeros((n_in,n_out), dtype=theano.config.floatX), name='delta_W_rec') be: self.delta_W_rec = theano.shared(value = numpy.zeros((n_out,n_out), dtype=theano.config.floatX), name='delta_W_rec') ?
in `https://github.com/roatienza/Deep-Learning-Experiments/blob/master/Experiments/Tensorflow/GAN/dcgan_mnist.py` you compute the generator loss as: `a_loss = self.adversarial.train_on_batch(noise, y)` but this also trains the discriminator using only the fake samples. shouldn't you freeze the discriminator weights for...
i am getting weird flickering when trying to apply engrave to a video `gmic.run('input /content/input.mp4 fx_engrave 0.5,50,0.8,40,0,0,0,10,1,0,0,0,1,0,50,50 output /content/gmic.mp4,24.800000')` what i tried: - made sure my code works ok when...
would be useful to have the astype=np.uint8 also do the *255 beforehand. otherwise using this option will return 0. not sure what is the convention in other libraries, but at...
from_numpy and to_numpy take care of channel interleaving, and from_numpy (as well as from_numpy_helper) work as expected with assuming numpy dim order is y,x and gmic is x,y. however i...
running on colab i see random crashes when i use from_numpy or to_numpy. the colab crashes without any message so its hard to debug. when from_numpy_helper(im, deinterleave=True), to_numpy_helper(im, interleave=True) it...
i am getting much worse results with default settings when running on video vs using ffmpeg and running on frames this is my code: https://colab.research.google.com/github/eyaler/avatars4all/blob/master/gmic_video.ipynb you can see both examples...
how can i programmatically get a list of all available features for my gmic-py installation?
another bug by myself: `out_images, _ = zip(*sorted(zip(out_images, scores), key=lambda x: -x[1]))` should be `out_images, _ = zip(*sorted(zip(out_images, scores), key=lambda x: x[1]))` as these are ce (or ppl in latest...
You would want to remove use_cache=False and get the 10x speedup they implemented for use_cache=True... I used the latest code but failed to flip the switch. Sorry for this!