Your images achieved an average structural similarity index of 0.70. At least 0.88 is required to pass.
Detailed list:
structural similarity index for ‘salientcat1.jpg’: 0.74.
structural similarity index for ‘salientcat2.jpg’: 0.72.
structural similarity index for ‘salientcatanddog.jpg’: 0.77.
structural similarity index for ‘salientdog1.jpg’: 0.59.
structural similarity index for ‘salientdog2.jpg’: 0.65.
Tried several times, even after loading the weights for 95 epochs, not able to cross 0.70 in ssim.
did you check similar other post @sindhus ?
Yes, I looked through similar discussions for salience maps, but still couldn’t figure out the issue.
@Deepti_Prasad , i have dm’ed screenshots
@sindhu the codes you have send is week 4 assignment but you created another topic stating saliency s week3 assignment. Also until mentor ask you for codes please don’t keep sending code DMs.
Please confirm me in the sentence DM which week assignment it is, so I can respond appropriately.
1 Like
issues with course3 week4 assignment
for model, for last conviction layer, use softmax activation.
in model compile, loss should sparse_categorical_crossentropuy.
Also just to point out, I noticed your model and model compile is in one cell where as for the assignment we are working upon is mentioned separately. model compile has to be written after the do saliency codes. so if you have made any chances, make sure to get a fresh copy from the classroom page.
in the do saliency codes
the first code line where you read the images you are using incorrect variables. you have used image path where as you were suppose to image to read the image from the arguments mentioned in that grade cell.
then while converting the channel, please rename that step as img and not image (second step) as image is being used as argument in general for that code cell which can cause incorrect conversion in the next steps.
for resize, recall that step as img
for adding an additional dimension, recall it as images not image
next in one of the codes, you mentioned
gradient_color = gradient, I don’t know what it does here. it is not required.
let me know the result. Also as mentioned by the l.t. if you stil notice any submissions failure, you are suppose to try the fallback session step for version mismatch issues.
Thank you very much for reviewing my code Deepti! I will make the corrections and try again.