IOU value 0

[MelanieFo94]

Hi,
I tested the Custom Training with 7 Lines of Code program. I have a dataset with train and test folder as required. The train.json and test.json were generated without any problem.

I train with a custom dataset, just one class. I have adapted all the code following this github and this "tutorial" : https://towardsdatascience.com/custom-instance-segmentation-training-with-7-lines-of-code-ff340851e99b

I set 50 epochs to train with my custom dataset.

But when I evaluated my models, the IOU value was 0 for all my models.

Do you have any explanation ? Apparently i'm not the only one to have this problem.

[aminrabinia]

Hi, I tested the Custom Training with 7 Lines of Code program. I have a dataset with train and test folder as required. The train.json and test.json were generated without any problem.

I train with a custom dataset, just one class. I have adapted all the code following this github and this "tutorial" : https://towardsdatascience.com/custom-instance-segmentation-training-with-7-lines-of-code-ff340851e99b

I set 50 epochs to train with my custom dataset.

But when I evaluated my models, the IOU value was 0 for all my models.

Do you have any explanation ? Apparently i'm not the only one to have this problem.

I had the same problem and I think it resolved when I converted my images from png to jpg.

[ayoolaolafenwa]

Hi, I tested the Custom Training with 7 Lines of Code program. I have a dataset with train and test folder as required. The train.json and test.json were generated without any problem.

I train with a custom dataset, just one class. I have adapted all the code following this github and this "tutorial" : https://towardsdatascience.com/custom-instance-segmentation-training-with-7-lines-of-code-ff340851e99b

I set 50 epochs to train with my custom dataset.

But when I evaluated my models, the IOU value was 0 for all my models.

Do you have any explanation ? Apparently i'm not the only one to have this problem.

@MelanieFo94 PixelLib is having issue with the latest version of tensorflow(2.6.0). Downgrade to any of the tensorflow versions between 2.0-2.50.

[ayoolaolafenwa]

Hi, I tested the Custom Training with 7 Lines of Code program. I have a dataset with train and test folder as required. The train.json and test.json were generated without any problem.
I train with a custom dataset, just one class. I have adapted all the code following this github and this "tutorial" : https://towardsdatascience.com/custom-instance-segmentation-training-with-7-lines-of-code-ff340851e99b
I set 50 epochs to train with my custom dataset.
But when I evaluated my models, the IOU value was 0 for all my models.
Do you have any explanation ? Apparently i'm not the only one to have this problem.

I had the same problem and I think it resolved when I converted my images from png to jpg.

Thank you for your contribution.

[MelanieFo94]

Hi, I tested the Custom Training with 7 Lines of Code program. I have a dataset with train and test folder as required. The train.json and test.json were generated without any problem.
I train with a custom dataset, just one class. I have adapted all the code following this github and this "tutorial" : https://towardsdatascience.com/custom-instance-segmentation-training-with-7-lines-of-code-ff340851e99b
I set 50 epochs to train with my custom dataset.
But when I evaluated my models, the IOU value was 0 for all my models.
Do you have any explanation ? Apparently i'm not the only one to have this problem.

@MelanieFo94 PixelLib is having issue with the latest version of tensorflow(2.6.0). Downgrade to any of the tensorflow versions between 2.0-2.50.

Hi, Indeed I had the 2.6.0 Tensorflow version. I switched it to another version (2.2.0) and now it's work ! Thank you very much !