[Python] Mismatched number of tokens and features: LLaVA

model_id = "llava-hf/llava-1.5-7b-hf"
pipe = pipeline("image-to-text",
model=model_id,
model_kwargs={})

def test_single_image():
"""
Test the model with a single image from the dataset.
"""
# Load just one image
images_data = load_test_images(num_images=1)

if not images_data:
print("No images found in the dataset")
return

# Get the first image
image_data = images_data[0]
print(f"Testing with image {image_data['index']}")
print(f"URL: {image_data['image_url']}")
print("\nOriginal hierarchical caption:")
print(image_data['hierarchical_caption'])

# Download and process the image
image = get_image_from_url(image_data['image_url'])
if image is None:
print("Failed to download image")
return

print("\nGenerating caption...")
generated_caption = generate_hierarchical_caption(image, pipe)

if generated_caption is None:
print("Failed to generate caption")
return

print("\nGenerated hierarchical caption:")
print(generated_caption)
print("\nComparison:")
print("-" * 80)
print("Original:", image_data['hierarchical_caption'])
print("Generated:", generated_caption)


The exception:

ValueError: Image features and image tokens do not match: tokens: 1, features 576


I keep getting this error. Please help! Not sure what to do. I've been trying to generate a caption but keep running into this error. Has anyone else seen this error? Any ideas of what to do? I greatly appreciate it.

Continue reading...