I found doing linear interpolation in the latent space quite gratifying to watch. So I have created this respository which allows you to create a countdown or countup GIF with MNIST images using smooth transitions that comes from the latent space.
You can choose any range of numbers from 0-9 to create the GIF.
| Countdown 9-0 | Count Up 0-9 | Countdown 5-0 | Count Up 0-5 | Countdown 9-5 | Count Up 5-9 |
|---|---|---|---|---|---|
 |
 |
 |
 |
 |
 |
You need to first download this pre-trained model to begin generating your own GIFs.
Then simply run the following:
python main.py --model-path <path-to-saved-model>
You can find below a detailed list of the available command line arguments that you can use to customize your GIF.
| Long | Short | Description | Default |
|---|---|---|---|
| --train | -t | flag to train model from scratch | False |
| --epochs | -e | number of epochs used for training | 250 |
| --batch-size | -b | batch size used in training | 400 |
| --save-interval | -s | save model every interval | 50 |
| Long | Short | Description | Default |
|---|---|---|---|
| --generate | -g | flag to generate gif from an existing model | True |
| --imsize | -i | image size of generated gif | 128 |
| --start-num | -a | count down/up starts with this number | 9 |
| --end-num | -z | count down/up ends with this number | 0 |
| --n-steps | -n | number of steps used in latent interpolation | 30 |
| --out-path | -o | output folder path to save generated gif | ./output |
| --out-name | -l | name of generated gif | countdown.gif |
| --model-path | -m | path to pre-trained Keras model | models/model-250.h5 |