Float32Array byte length error when storing images in the cloud

[Edward MAN]

I’m working on an image recognition project and ran into an issue. When I go to the Train or Test page, I get this error:
“Byte length of Float32Array should be a multiple of 4.”

This only happens when I store the data in the cloud. If I store the images in the browser, everything works fine.

I’ve attached the images that appear to be causing the problem below.

Any idea what might be causing this or how to fix it? Thanks!

[Dale Lane]

Sorry to hear you've had problems with this. 

I'm afraid there aren't any images attached to your post, so it's difficult to see exactly what has gone wrong here. Could you give me the full URL for your cloud project please? That would give me enough information to get started.

Kind regards

D

[Edward MAN]

Hello, Dale. I have attached the images here. Please have a look. 
My training data: https://drive.google.com/file/d/1LIZYs5w5nVn_vzssMwK0yf3JIcKG6YKO/view?usp=sharing
The error message: https://drive.google.com/file/d/1bL8wBNZ6Y2jFUmloGgkQfKTDVIx0qrM6/view?usp=sharing

Best wishes,

Edward

[Dale Lane]

Sorry, that isn't really very helpful. 

If you could just give me the full URL for your cloud project, please, that would give me enough information to get started. 

Kind regards

D

[Edward MAN]

Sure. Please have the full URL of my cloud project:
https://machinelearningforkids.co.uk/#!/mlproject/auth0%7C60014b1b5dbf6e006e990543/c3629ee0-5707-11eb-95c4-7f9cd564af30

Thanks a lot,

Edward

[Dale Lane]

I'm not able to recreate the error, sorry - your project trained successfully for me. 

That suggests that the problem isn't with the training images, but something unique to your browser or computer. 

Please can you collect a log of what happens on your computer when you try to train the project. Instructions for how to do this can be found at https://machinelearningforkids.co.uk/#!/debug 

(For step 3, you would click on the "Train new machine learning model" button and wait for the error to be displayed)

If you can send me the log, this will contain a record of everything that happens leading up to the error, which will hopefully give me a clue what the source of the problem is. 

Kind regards

D

[Edward MAN]

Hello Dale,

I’ve recreated the error and attached the log to the thread. The error appeared shortly after I landed on the Train & Test page. The gear animation kept running, and the “Train new machine learning model" button wasn't shown.

Please have the log file here:
https://drive.google.com/file/d/1MCPllN0Ndk5qlM8NZXn8zsDkkMdtLIvV/view?usp=sharing

Thanks,

Edward

[Dale Lane]

When you go to the page, a mobilenet model is downloaded and prepared for use as a base for training your custom model. (When you click on the button to train a new model, this pretrained mobilenet model is used as the basis for transfer learning)

You can see the model being downloaded in the Network tab of your browser's developer console, split up into 56 files which are each downloaded separately and then assembled on your computer. 

These files are cached (stored) on your computer, so that you don't need to wait for this to download every single time - the first time you access the page, the files are downloaded, subsequent times they're just loaded from disk. 

The log suggests that one (or more) of the model files is corrupt. Presumably there was some error during download, and instead of getting a model file, something else invalid was stored. 

If this is correct, the fix is to re-download these files to get a correct valid copy of them. 

You can do this by clearing or disabling your browser cache, which will force your browser to download them again the next time you visit the page.

Kind regards

D

[Edward MAN]

Thank you, Dale. The problem was solved.