Rasmus Lindahl

icon indicating an email [email protected]

icon location Sweden icon location Fascinated by most things. I like to experiment and test the limits of intuition. Other than that I like to read, play the piano and help people.

Results 16 comments of Rasmus Lindahl

error: command 'x86_64-linux-gnu-gcc' failed with exit status 1 on google colab

Any updates?

Results

Are you inputting descriptions for images or just let it randomly generate an image?

Results

Would you mind sharing the Colab you have so far? :)

Results

Thanks! I'm a noob but I tried to help: https://colab.research.google.com/drive/1KxG1iGBoKt2fLVH7uXG_vhvll2OlFkey?usp=sharing :)

Results

Okay. Here is a fully working Colab for at least VAE training. Thanks to NaxAlpha of course! https://colab.research.google.com/drive/1KxG1iGBoKt2fLVH7uXG_vhvll2OlFkey?usp=sharing ![image](https://user-images.githubusercontent.com/32551374/104041571-8cfbd700-51d9-11eb-8fd4-707ddeab5658.png) After around 600 training pairs.

Results

Which dataset are you using to train DALL-E? Don't you need text as well? Also, what are you training on? Do you have access to Google Colab Pro?

Results

Would it possible that using more coherent text (instead of random) also would result in more coherent images?

Results

@NaxAlpha I created an image + desc fetcher. You can see it [here](https://github.com/lucidrains/DALLE-pytorch/issues/13#issue-782604959). Could it be useful?

Results

I just benchmarked my scraper on Google Colab Pro. It takes around 3.46 hours/10 000 image+desc pairs. I will upload the data when I'm done.

Results

Nevermind my scraper. Just use the COCO dataset. It has 500 000 images with descriptions for each one. Takes 10 minutes to download on Colab Pro. If anyone wants me...