Hello, thanks for sharing this implementation. I'm interested in the complementary search/recommendations for fashion but I don't understand how to create the embeddings for inference.
First of all, can you confirm if it is needed to download all of these (download_data.py)?
if __name__ == "__main__":
get_images_stl("fashion", "product")
get_images_complete_the_look("fashion_v2", "train")
get_images_complete_the_look("fashion_v2", "test")I suppose get_images_stl is to download the fashion complementary data from STL, right? Do I need to download the other two (fashion_v2 train/test) if I just want to make inference for complementary fashion items?
To generate the embeddings for compatible products you say:
Get compatible product embedding - Run cd src, make sure the in the
features/Embedding.py, the class method similar_product_embedding
is being selected, then run PYTHONPATH=../:. python features/Embedding.py
(be careful this could take up to 15 hours without the presence of a GPU, 7 hours
with GPU)<img
Which one of the methods should I need to run just for the complementary search/recommendation?
if __name__ == "__main__":
StyleEmbedding().similar_product_embedding(
data_loader=FashionProductSTLDataloader().data_loader(), task_name="similar_product"
)
StyleEmbedding().similar_product_embedding(
data_loader=FashionCompleteTheLookDataloader().single_data_loader(),
task_name="similar_prod_CTL",
)
StyleEmbedding().compatible_product_embedding(
data_loader=FashionCompleteTheLookDataloader(image_type="test").single_data_loader(),
task_name="compatible_product_test",
)
StyleEmbedding().compatible_product_embedding(
data_loader=FashionCompleteTheLookDataloader().single_data_loader(),
task_name="compatible_product",
)Thanks!
Hello, thanks for sharing this implementation. I'm interested in the complementary search/recommendations for fashion but I don't understand how to create the embeddings for inference.
First of all, can you confirm if it is needed to download all of these (
download_data.py)?I suppose
get_images_stlis to download the fashion complementary data from STL, right? Do I need to download the other two (fashion_v2 train/test) if I just want to make inference for complementary fashion items?To generate the embeddings for compatible products you say:
Which one of the methods should I need to run just for the complementary search/recommendation?
Thanks!