Everything works well until I wanted to save it.
My results are different than one that I saw on video.
video
mine
why our shapes are different?
Thank you
@oduvan : I dont recollect exactly how it worked in the video. But, In your case, you have 5 images for each test sample. This is achieved by augmentation (default n_aug=4 ) +1 for original image. Usually we average the predictions give the result as the prediction Hence the shape discrepancy that you see. I believe jeremy would have averaged the predictions across the 1st axis (i.e. across 5 samples/test image) to obtain the result.
~Gokkul
thank you for your answer.
how can I get averaged the predictions across the 1st axis?
You should compute mean along axis 0 i.e "preds = np.mean(np.exp(log_preds),0) "