Invalid annotations

[dev2y]

I am trying to train a model with bigger images (2000x2000x3). I have enough compute and memory and I am using the following training command

python train.py --snapshot-path /TrainingOutput/snapshots --image-min-side 2000 --image-max-side 2000 --no-resize --epochs --random-transform --steps 12 pascal /TrainingData/

I get the following warning.

TrainingData/JPEGImages/W09_6.jpg with id 8 (shape (2000, 2000, 3)) contains the following invalid boxes: [[ -1. 65. 44. 107.]].

I did check my annotation xml file and I don't have this box. Instead I have this.

        <bndbox>
            <xmin>785</xmin>
            <ymin>65</ymin>
            <xmax>824</xmax>
            <ymax>94</ymax>
        </bndbox>

I am wondering if script is adjusting box to new image size, though I said no-resize.

Thanks\

[dev2y]

I reduced the images to 800x800 but I still get the same error. I double checked and I don't see any negative number in xml files. I did try with --no-resize option. No luck yet. Please let me know if some one has seen this error and found a solution.

[hlmhlr]

Dear @dev2y, how did youmanage to resize the images by maintaing the same sizes of bounding boxes of objects as well? Though this is not the solution towards your problem but i would be grateful if you could help me in any way.

Thanks,

[dev2y]

@hlmhlr, we didn't resize. Instead we cropped smaller images from bigger images. Sorry. I should have been more clearer. We wrote script to create new annotation files for the cropped images.

[hlmhlr]

@hlmhlr, we didn't resize. Instead we cropped smaller images from bigger images. Sorry. I should have been more clearer. We wrote script to create new annotation files for the cropped images.

@dev2y , Ah okay, thanks for your response. I need to look for some other ways.

[stale[bot]]

This issue has been automatically marked as stale due to the lack of recent activity. It will be closed if no further activity occurs. Thank you for your contributions.