engindeniz / vitis Goto Github PK

Great work!
Can you provide trained weights and code to run inference locally on my own videos?

First of all, thanks to your great work!

I wonder about the arguments related with "visual prompts".
To run your code on custom dataset I'm preparing to extract features.

When I tried to extract visual features, i got some troubles.

self.encoder = CLIPEncoder(config, self.args.prompt_type, self.args.visual_prompt_projection_factor,
AttributeError: 'Namespace' object has no attribute 'visual_prompt_projection_factor'

Actually, not only about 'visual_prompt_projection_factor'.
I also can't find about these informations in your args file.

        self.encoder = CLIPEncoder(config, self.args.prompt_type, self.args.visual_prompt_projection_factor,
                                   self.args.visual_prompt_projection_layer, self.args.visual_model_hidden_size,
                                   self.args.visual_prompt_num_layer, self.args.visual_prompt_num_tokens,
                                   self.args.visual_prompt_initialization, self.args.visual_prompt_patch_size,
                                   self.args.visual_prompt_dropout_rate, self.args.visual_prompt_like_linear_layer_each_transformer_layer)

And when extract the features, I think we would not use promts and projections right?
Would you help me to shoot this troubles?

I think this will affect to follow-up tasks like training and testing.