Convolutional Auto-encoder trained on MSCOCO + Flickr 30k. Total ~ 200k images
The model works with both jpg and png images.
To retrain the model: python3 train_encoder.py
If you want to retrain the model, do not forget to put your images in 0 folder in images. You need TFlearn and Keras
To evaluate the model: python3 evaluate.py <checkpoints/checkpointname> <path_to_image>
If you clone this repo, put model.h5 instead of checkpoint_name
The output image by default will have jpg extension. You can change that in evaluate.py
JPG image example:
PNG image example:
References:
Original Github repo