How to colorize without giving the hints, just pass the grey image and get the colored image?

[mushtaqpatel0505]

How to colorize without giving the hints, just pass the grey image and get the colored image

[mushtaqpatel0505]

@YeolJ00 @niceDuckgu @pmh9960

[YeolJ00]

Our model is not specifically trained for unconditional colorization, but you can still obtain a colorized image without providing any hints. Have you tried it in the demo or the code? You simply don't pass any hints.

[mushtaqpatel0505]

Yeah, for some images it's working well (colorizing the grey image) but for some images, it is showing the grey image as it is for old family photos it's not working well

[mushtaqpatel0505]

Do I need to train on the old family photo? Could you tell me the procedure for using current checkpoints? base, small or tiny

[YeolJ00]

For some images, the initial results may still be grey since it is not trained to do that. If you with to train an unconditional model, you can use the following command, given that you have all the training data ready.

bash scripts/train.sh --num_hint_range 0

Also, in order to use the small and tiny model, you would have to change the model type with the following argument.

bash scripts/infer.sh --model icolorit_small_4ch_patch16_224 --model path/to/your/checkpoint

[mushtaqpatel0505]

Could I train the "icolorit_base_4ch_patch16_224" checkpoint further on the coco, place365 ? not from scratch

[YeolJ00]

Yes you can, simply use the --resume argument and load the pretrained model. I strongly suggest reading through the arguments in main.py.

[mushtaqpatel0505]

How to run on single image?

[YeolJ00]

Please refer to the demo for a GUI. Also, PLEASE provide information when posting a question with information related to what you have tried, what the errors were and the enviornent you are running on.

[mushtaqpatel0505]

How to run on the single image without GUI in the terminal/command prompt? input: grey image output: colored image