Yolov3: Label path issues when training on own dataset

[YxChen98]

I was following this instruction to train custom dataset:
https://github.com/ultralytics/yolov3/wiki/Train-Custom-Data

And I have the dataset downloaded by running get_coco_dataset.sh

I tried to use the given coco_1cls.data as the training data with the given yolov3-1cls.cfg as the cfg file
My input command was:
python train.py --data data/coco_1cls.data --cfg cfg/yolov3-1cls.cfg

And what I got was this:
Traceback (most recent call last):
File "train.py", line 341, in
multi_scale=opt.multi_scale,
File "train.py", line 147, in train
multi_scale=multi_scale)
File "F:\yolo\nest\utils\datasets.py", line 208, in __init__
assert len(np.concatenate(self.labels, 0)) > 0, 'No labels found. Incorrect label paths provided.'
AssertionError: No labels found. Incorrect label paths provided.

I am using PyCharm 2019,1,2 on Windows 10

Has anyone experienced this issue?

[glenn-jocher]

@YxChen98 the error message is self-explanatory: No labels found. Incorrect label paths provided

[glenn-jocher]

The command works perfectly if you set up your environment correctly.
https://colab.research.google.com/drive/1G8T-VFxQkjDe4idzN8F-hbIBqkkkQnxw#scrollTo=R51pP88Rj02x

!python train.py --data data/coco_1cls.data --cfg cfg/yolov3-1cls.cfg

Namespace(accumulate=4, backend='nccl', batch_size=16, cfg='cfg/yolov3-1cls.cfg', data_cfg='data/coco_1cls.data', dist_url='tcp://127.0.0.1:9999', epochs=68, evolve=False, img_size=416, multi_scale=False, nosave=False, notest=False, num_workers=4, rank=0, resume=False, transfer=False, var=0, world_size=1)
Using CUDA device0 _CudaDeviceProperties(name='Tesla T4', total_memory=15079MB)

Model Summary: 222 layers, 6.15237e+07 parameters, 6.15237e+07 gradients

   Epoch       Batch        xy        wh      conf       cls     total   targets      time
    0/67         0/0     0.238     0.342      49.3         0      49.8        17      3.67
Reading image shapes: 100% 5/5 [00:00<00:00, 7332.70it/s]
               Class    Images   Targets         P         R       mAP        F1
Computing mAP: 100% 1/1 [00:01<00:00,  1.73s/it]
                 all         5        17         0         0         0         0

   Epoch       Batch        xy        wh      conf       cls     total   targets      time
    1/67         0/0     0.215     0.391      49.3         0      49.9        15      8.23
               Class    Images   Targets         P         R       mAP        F1
Computing mAP: 100% 1/1 [00:00<00:00,  1.02it/s]
                 all         5        17         0         0         0         0
...

[YxChen98]

I got it solved.
The coco_1cls.txt needs some change:
I changed the previous
../coco/images/val2014/COCO_val2014_000000001464.jpg
to
./data/images/val2014/COCO_val2014_000000001464.jpg
and it works

[MrDavinci]

Hello, I got the same problem under win10.
Chen, I gusse you put the images and labels into the /data. And I did the same thing as you.
那么老哥可以加个qq嘛, save me pls.

[glenn-jocher]

@MrDavinci in the custom training tutorial yolov3 and coco and folders at the same level, they exist next to each other.
https://github.com/ultralytics/yolov3/wiki/Train-Custom-Data

[logic03]

I caught the same question one hours ago, and I finally find the reason is that my picture path in data/train.txt & data/collector.name & data/collector.data is relative path, I changed it to absolute path like names=C:\Users\27319\Documents\GitHub\yolov3-master\data\collector.names, the problem would be solved!

[logic03]

I caught the same question one hours ago, and I finally find the reason is that my picture path in data/train.txt & data/collector.name & data/collector.data is relative path, I changed it to absolute path like names=C:\Users\27319\Documents\GitHub\yolov3-master\data\collector.names, the problem would be solved!