Function compute_clip_embeddings

utils/poster_eval_utils.py:240–272

Loads each image in `folder`, encodes it with the CLIP model, and returns a list (or array) of embeddings, shape (N, D).

(folder, model, processor, device)

def compute_clip_embeddings(folder, model, processor, device):
    """
    Loads each image in `folder`, encodes it with the CLIP model,
    and returns a list (or array) of embeddings, shape (N, D).
    """
    model.eval()
    embeddings = []

    # Gather all image files
    image_files = [
        f for f in os.listdir(folder)
        if f.lower().endswith(('.png', '.jpg', '.jpeg'))
    ]

    if not image_files:
        print(f"No valid images found in {folder}")
        return np.array([])

    for filename in image_files:
        img_path = os.path.join(folder, filename)
        image = Image.open(img_path).convert('RGB')

        # Preprocess for CLIP
        inputs = processor(images=image, return_tensors="pt").to(device)

        # Encode and get the image embeddings
        with torch.no_grad():
            clip_emb = model.get_image_features(**inputs)
            # Move to CPU and convert to NumPy
            clip_emb = clip_emb[0].cpu().numpy()
            embeddings.append(clip_emb)

    return np.array(embeddings)  # shape: (N, D)

def compute_clip_embedding(input_data, model, processor, device='cuda', input_type=None):
    """