Inference request

[nancvcs]

Thank you for such a detail repo. I have successfully reimplemented it and got the test accuracy 66.7% (I'm happy with that because my dataset is quite big). I want to inference a new single test image to get the most similar image with it (based on lowest contrastive loss) but the result is not similar to the test image at all, even though the test image is compared with just some folders of image, not the whole train dataset, and even though I copied the test image to one of the folders. Can you have a look at my code below? Thank you.

reference_loss = []
pairs_infer = []
img1 = process_image(img_path=test_path)
for i in filename:
  img2 = process_image(img_path=i)
  pairs_infer.append([img1, img2])

y_pred_te = model.predict([pairs_infer[:, 0], pairs_infer[:, 1]])

most_similar_index = np.argmin(y_pred_te)
most_similar_image_path = filename[most_similar_index]