Evaluate custom and HuggingFace text-to-image/zero-shot-image-classification models like CLIP, SigLIP, DFN5B, and EVA-CLIP. Metrics include Zero-shot accuracy, Linear Probe, Image retrieval, and KNN accuracy.
Now, when executing the CLI's animation command each target model will be individually selected, thus avoiding the pitfall where the user mistakenly selected an amount of models different than 2.
Also, added an improved naming for the animation file to avoid the repetition of the dataset name and also make it more "portable" to Windows OS (previous date used the : character which is invalid in Windows filesystem).
Now, when executing the CLI's animation command each target model will be individually selected, thus avoiding the pitfall where the user mistakenly selected an amount of models different than 2.
Also, added an improved naming for the animation file to avoid the repetition of the dataset name and also make it more "portable" to Windows OS (previous date used the
:
character which is invalid in Windows filesystem).