AssertionError: non-normalized or out of bounds coordinate labels:

[Samjith888]

Got the following error while running for custom data with 7 classes. https://github.com/ultralytics/yolov3/issues/569 .

$ python train.py --data data/coco.data
Namespace(accumulate=2, adam=False, arc='default', batch_size=32, bucket='', cache_images=False, cfg='cfg/yolov3-spp.cfg', data='data/coco.data', device='', epochs=273, evolve=False, img_size=416, img_weights=False, multi_scale=False, name='', nosave=False, notest=False, prebias=False, rect=False, resume=False, transfer=False, var=None, weights='weights/ultralytics49.pt')
Using CUDA device0 _CudaDeviceProperties(name='GeForce GTX 1070', total_memory=8116MB)
AssertionError: non-normalized or out of bounds coordinate labels:
Reading labels: 0%| | 0/1562 [00:00<?, ?it/s]Traceback (most recent call last):
File "train.py", line 444, in
train() # train normally
File "train.py", line 193, in train
cache_images=False if opt.prebias else opt.cache_images)
File "/home/samjth/Desktop/yolov3/utils/datasets.py", line 335, in init
assert (l[:, 1:] <= 1).all(), 'non-normalized or out of bounds coordinate labels: %s' % file
AssertionError: non-normalized or out of bounds coordinate labels: /home/samjth/Desktop/yolov3/data/coco/labels/train2014/073m_WW_5049.txt
Reading labels: 0%| | 0/1562 [00:00<?, ?it/s]
(base)

[FranciscoReveriano]

Your labels are not in the correct coco-format.

[Samjith888]

Your labels are not in the correct coco-format.

Do you meant about the directory structure of coco or the label.txt file ?

Label.txt is following format

Class_id Xmin Ymin width height

[glenn-jocher]

@Samjith888 see https://docs.ultralytics.com/yolov5/tutorials/train_custom_data

Each row is class x_center y_center width height format, in normalized coordinates (0-1).

[Samjith888]

@Samjith888 see https://docs.ultralytics.com/yolov5/tutorials/train_custom_data

Each row is class x_center y_center width height format, in normalized coordinates (0-1).

My label.txt is like following
class x_center y_center width height
2 2205.499392 0.0 576.5570559999996 2160.0

And the image resolution is 4096 *2160

Through normalizing the x_center y_center width height values, we can convert it into 0-1 range values. While training, how this model will locate our ROI from the 4096 *2160 resolution image by using the normalized values (between 0-1)??

[glenn-jocher]

@Samjith888 we've updated the Custom Training Example in the wiki with directions for normalizing. All labels must be normalized otherwise training will not work.

[Samjith888]

@Samjith888 we've updated the Custom Training Example in the wiki with directions for normalizing. All labels must be normalized otherwise training will not work.

Values before normalizing
x_center y_center width height
2029.5434240000002 1397.85804 2257.0106880000003 1524.2839199999999

After Normalization
0.9396034370370371 0.341273935546875 1.0449123555555557 0.37213962890624996

I have followed the instructions in the wiki for normalizing the values. And the image resolution is 4096 2160 (wh). But i got the width value is greater than one (1.0449123555555557) even after normalization.

[glenn-jocher]

@Samjith888 well maybe your image width and height are switched, or your original coordinates are xyxy rather than xywh.

[Samjith888]

@Samjith888 well maybe your image width and height are switched, or your original coordinates are xyxy rather than xywh.

Getting the following error after solving the above error

Downloading https://drive.google.com/uc?export=download&id=1uTlyDWlnaqXcsKOktP5aH_zRDbfcDp-y as yolov3.weights... Done (147.5s) Traceback (most recent call last): File "train.py", line 444, in <module> train() # train normally File "train.py", line 131, in train cutoff = load_darknet_weights(model, weights) File "/home/samjth/Desktop/yolov3/models.py", line 352, in load_darknet_weights conv_w = torch.from_numpy(weights[ptr:ptr + num_w]).view_as(conv_layer.weight) RuntimeError: shape '[1024, 512, 3, 3]' is invalid for input of size 1891940

I have 7 classes ,hence filter= 36. I have changed the classes and filter values in 3 fields of the yolov3.cfg file.

[glenn-jocher]

@Samjith888 you may want to try loading the pytorch weights instead --weights yolov3.pt or no pretrained weights at all --weights '' since it seems that your cfg is not compatible with yolov3.weights.

[Damon2019]

@Samjith888 hi ,have you succeed?

[HassanBinHaroon]

@Damon2019 @Samjith888

This problem arises when the bounding box coordinate(s) is(are) greater than the respective image dimension(s).

For Instance: My image dimensions are 640 (width) x 512 (height) and due to some reason, my bounding box coordinates are xmin = 600, ymin = 450, xmax = 660, ymax = 510.

In the above example it can be seen that xmax > 640 and after normalizing, the result wouldn't be between 0-1. It would be greater than 1. This kind of situation is the cause of this error.