A question about how to test retrieval

[Jinpeng Wang]

Hi, thank you for building such a useful and concise framework for image retrival task!

I just read the file inference.py and got confused about the testing process.

PyTorch-Image-Retrieval/inference.py

Lines 20 to 23 in 8116251

 query_loader = DataLoader(query_img_dataset, batch_size=infer_batch_size, shuffle=False, num_workers=4, 
                           pin_memory=True) 
 reference_loader = DataLoader(reference_img_dataset, batch_size=infer_batch_size, shuffle=False, num_workers=4, 
                               pin_memory=True) 

My question is, why is the database for retrieval getting divided into batches? Shouldn't we keep it as a whole for every query, since we want to rank all the reference images in whole database according to the pair-wise similarity?

[Sangwon Lee]

In the batch_process function, all feature vectors are concatenated.

PyTorch-Image-Retrieval/inference.py

Lines 28 to 29 in 8116251

 query_paths, query_vecs = batch_process(model, query_loader) 
 reference_paths, reference_vecs = batch_process(model, reference_loader) 

PyTorch-Image-Retrieval/inference.py

Lines 111 to 122 in 8116251

 def batch_process(model, loader): 
     feature_vecs = [] 
     img_paths = [] 
     for data in loader: 
         paths, inputs = data 
         feature_vec = _get_feature(model, inputs.cuda()) 
         feature_vec = feature_vec.detach().cpu().numpy()  # (batch_size, channels) 
         for i in range(feature_vec.shape[0]): 
             feature_vecs.append(feature_vec[i]) 
         img_paths = img_paths + paths 
  
     return img_paths, np.asarray(feature_vecs)