mirror of
https://github.com/khoaliber/khoj.git
synced 2026-03-09 05:39:12 +00:00
Batch encode XMP metadata from images too for image_search
This commit is contained in:
@@ -61,8 +61,10 @@ def compute_embeddings(image_names, model, embeddings_file, batch_size=50, regen
|
|||||||
print(f"Saved computed embeddings to {embeddings_file}")
|
print(f"Saved computed embeddings to {embeddings_file}")
|
||||||
|
|
||||||
if image_metadata_embeddings is None:
|
if image_metadata_embeddings is None:
|
||||||
image_metadata = [extract_metadata(image_name, verbose) for image_name in image_names],
|
image_metadata_embeddings = []
|
||||||
image_metadata_embeddings = model.encode(image_metadata, batch_size=batch_size, convert_to_tensor=True, show_progress_bar=True)
|
for index in trange(0, len(image_names), batch_size):
|
||||||
|
image_metadata = [extract_metadata(image_name, verbose) for image_name in image_names[index:index+batch_size]]
|
||||||
|
image_metadata_embeddings += model.encode(image_metadata, convert_to_tensor=True, batch_size=batch_size)
|
||||||
torch.save(image_metadata_embeddings, f"{embeddings_file}_metadata")
|
torch.save(image_metadata_embeddings, f"{embeddings_file}_metadata")
|
||||||
if verbose > 0:
|
if verbose > 0:
|
||||||
print(f"Saved computed metadata embeddings to {embeddings_file}_metadata")
|
print(f"Saved computed metadata embeddings to {embeddings_file}_metadata")
|
||||||
|
|||||||
Reference in New Issue
Block a user