This is the starter repository for the Weights & Biases community benchmark for predicting the behavior of catz from a dataset of cat GIFs.

1. Be sure to sign up for W&B.
2. Clone this repository: git clone https://github.com/wandb/catz.git
3. Run pip install -U -r requirements.txt to install requirements.
4. Run python train.py to train the baseline model. Modify this file and the data pipeline (or write your own scripts and create different model architectures!) to get better results.
5. Submit your results to the benchmark.

The dataset is comprised of sequences extracted from GIFs of cats thanks to GIPHY! Each cat has its own directory, which contains a sequence of 6 images. There are 6421 sequences in the training set and 1475 in the test set. Each image is 96x96 pixels.

The goal is to predict the 6th frame given 5 consecutive previous frames.

We use a perceptual distance metric (val_perceptual_distance) on the validation set to rank results (lower values are better).

You can submit your best runs to our benchmark. More specifically, go the "Runs" table in the "Project workspace" tab of your project. Hover over the run's name, click on the three-dot menu icon that appears to the left of the name, and select "Submit to benchmark".

• Use an RNN
• Different loss functions
• Data augmentation

Participating from Qualcomm? See this README for more details.