[Python] why there is no grad in the fine-tuning SAM

I download the SAM model and parameter from the official website and then try to re-train the network. When I try to do the back-propagation, then there is an error called:

RuntimeError: element 0 of tensors does not require grad and does not have a grad_fn
​

The training network is as:

url = "https://dl.fbaipublicfiles.com/segment_anything/sam_vit_b_01ec64.pth"
checkpoint_path = "sam_vit_b_01ec64.pth"

if not os.path.exists(checkpoint_path):
with requests.get(url, stream=True) as r:
r.raise_for_status()
with open(checkpoint_path, 'wb') as f:
for chunk in r.iter_content(chunk_size=8192):
f.write(chunk)
print("Checkpoint downloaded.")

model_type = "vit_b"
checkpoint = "sam_vit_b_01ec64.pth"

model = sam_model_registry\[model_type\](checkpoint=checkpoint)


Is there someone can help. Thanks a lot loll.

Continue reading...