How to generate .h5 Files for eval_goldfish_movie_chat.py

[prote376]

def load_frames(self, movie_name):
    filenames = sorted(os.listdir(os.path.join(self.dataset_path, movie_name)))
    
    filenames.sort(key=numerical_sort_key)
    # define torch tensor to store the frames of size(0,0,0)
    data = []
    for filename_number in tqdm(filenames,desc="Loading frames"):
        file_path = os.path.join(self.dataset_path, movie_name, filename_number)

        if not os.path.isfile(file_path):
            print(f"Did not find file: {filename_number}")
        try:
            with h5py.File(file_path, 'r') as h5_file:
                image_embeds=torch.tensor(h5_file[f"frames_{filename_number[:-3]}"][:])
                image_embeds = image_embeds[:,1:,:] # remove the first token (CLS) (200,256,1408)
                # concate each 4 neighbours image tokens 
                bs, pn, hs = image_embeds.shape
                image_embeds = image_embeds.view(bs, int(pn/4), int(hs*4)) 
                data.extend(image_embeds)

        except Exception as e:  
            print(f"Failed to process {filename_number}: {e}")
        
            
    frames=torch.stack(data)
    return frames

I was testing the code in eval_goldfish_movie_chat.py.
In the load_frames function of MovieChatDataset (Lines 112–137), it appears that the code expects .h5 files corresponding to each video file.
Could you please provide more details on how these .h5 files are generated?