search

hanzhanggit / StackGAN

MIT License
1.86k stars 457 forks source link

stage-I training problem #56

Open 18398639574 opened 5 years ago

18398639574 commented 5 years ago

Please switch to tf.summary.FileWriter. The interface and behavior is the same; this is just a rename. Traceback (most recent call last): |ETA: --:--:-- File "stageI/run_exp.py", line 78, in algo.train() File "/data/sdv1/WLL/StackGAN/stageI/trainer.py", line 346, in train num_embedding) File "/data/sdv1/WLL/StackGAN/misc/datasets.py", line 144, in next_batch sampled_images = self.transform(sampled_images) File "/data/sdv1/WLL/StackGAN/misc/datasets.py", line 78, in transform images[i][w1: w1 + self._imsize, h1: h1 + self._imsize, :] TypeError: slice indices must be integers or None or have an index method

akhilvasvani commented 5 years ago

Change images[i][w1: w1 + self._imsize, h1: h1 + self._imsize, :] --> images[i][int(w1): int(w1 + self._imsize), int(h1): int(h1 + self._imsize), :]

adigol commented 5 years ago

It runs 6 hours and doesn't stop, is it OK?

akhilvasvani commented 5 years ago

Well, that depends on what you're looking for. What is your batch size?

adigol commented 5 years ago

I ran the original code on the birds data(about 600 batches). I want to run it on the original data and then I'll run it on my data.

akhilvasvani commented 5 years ago

600 batches? I think you mean Epochs.

For me, it took about a day to train with a batch size of 64 on my NVIDIA 1080 Ti GPU.

If you want to train StackGAN on original data (both images and captions), you need to create your own char-CNN-RNN or skip-thoughts embeddings

priyanka1220 commented 1 year ago

can anyone please provide me the weights of stage1