Browser-based Models with TensorFlow.js C1_W4 assignment

Course Q&A TensorFlow: Data and Deployment Specialization
1

Have a problem, when using ‘https://storage.googleapis.com/tfjs-models/tfjs/mobilenet_v1_0.25_224/model.json’ model for the assignment the model trains and I can predict and save the model, but when I submit it flatten_Flatten1_input should have shape (7, 7, 256) but got array with shape (7, 7, 1024).
When I use the suggested model ‘https://storage.googleapis.com/tfjs-models/tfjs/mobilenet_v1_1.0_224/model.json’ the model in training gets an error WebGL:

‘’‘CONTEXT_LOST_WEBGL: loseContext: context lost
uncaught (in promise) Error: Failed to compile fragment shader.’‘’

and I can neither predict nor save the model for submission, what should I do in this case?
Basically the problem is the mobilenet model itself, because if I use another everything works perfectly fine, just the grader is unhappy!!!

2

Hello Roman,

I am also facing the same issue. Have you solved the above issue? Let me know your workaround for this issue.

Regards,
Uma

1 Like
3

Hi, only solution that worked for me was doing it from linux, tried other browsers in windows, had the same problem, you could use virtual box and install linux on it, or install linux on separate disk partition.

Hope it helps, good luck!

1 Like
4

Thanks for the suggestion. I will try that.

1 Like