Update parser options to take in any CLIP model + with JIT. The render image resolution is updated according to the model specifications.
Note: as new CLIP models become public, for the sake of efficiency it will probably be good to just replace the render preprocessing (e.g. setting the render resolution) with the preprocess function given in the CLIP model loading. We can just set the default render resolution to be the current max of the models to prevent too much blurring from upsampling.
Results shown on the ninja with each model with the same options + seed:
RN50
Update parser options to take in any CLIP model + with JIT. The render image resolution is updated according to the model specifications. Note: as new CLIP models become public, for the sake of efficiency it will probably be good to just replace the render preprocessing (e.g. setting the render resolution) with the
preprocessfunction given in the CLIP model loading. We can just set the default render resolution to be the current max of the models to prevent too much blurring from upsampling.Results shown on the ninja with each model with the same options + seed: RN50
RN101
RN50x4
RN50x16
RN50x64
VIT-B/32
VIT-B/16
VIT-L/14
VIT-L/14@336px