Darknet: Error in load_data_detection() - OpenCV when training on my custom data in windows 10

[asdsdlgl]

I have the issue when i train my custom data in windows 10, it will show the error " Error in load_data_detection() - OpenCV " when loading the image. And if i just test data using pretrained-weight not training, it will correctly detect the result.
And i try different versions of opencv, including 3.4.5 3.4.10 4.1.0, all of them are failed. My cuda version is 10.0 and vs2017.
How to resolve this problem that " Error in load_data_detection() - OpenCV " when training custom data?

[sharoseali]

HI, @asdsdlgl did you manage this problem. I am also stucked here... please help... All my images and their annotations are are in the correct path, but still don't know why its happening.. @AlexeyAB
I am using this command darknet.exe detector train data/drone.data cfg/yolov4-tiny_drone.cfg yolov4-tiny.conv.29 and results are like this

[MohamedAdelNaguib]

I have the issue when i train my custom data in windows 10, it will show the error " Error in load_data_detection() - OpenCV " when loading the image. And if i just test data using pretrained-weight not training, it will correctly detect the result.
And i try different versions of opencv, including 3.4.5 3.4.10 4.1.0, all of them are failed. My cuda version is 10.0 and vs2017.
How to resolve this problem that " Error in load_data_detection() - OpenCV " when training custom data?

Hi , I am also stuck in this error anyone have a solution?

[sharoseali]

OK, the error is resolved when I again generate the train and valid.txt files by writing the absolute path of all image files. Here is the code:

from os import listdir
from os.path import isfile, join
customPath = 'Your/abs/path/' 
onlyfiles = [f for f in listdir(customPath + '/DET-val/data/') if isfile(join(customPath + '/DET-val/data',f))]


trainFile = customPath + "/val.lists"
file = open(trainFile, 'w')


counter = 0
customPath = 'Your/abs/path/' 
for eachFile in onlyfiles:
    if "jpg" in eachFile:
        counter+=1
        file.write(customPath + eachFile + "\n")


print(counter)

[ganapathi12]

OK, the error is resolved when I again generate the train and valid.txt files by writing the absolute path of all image files. Here is the code:

from os import listdir
from os.path import isfile, join
customPath = 'Your/abs/path/' 
onlyfiles = [f for f in listdir(customPath + '/DET-val/data/') if isfile(join(customPath + '/DET-val/data',f))]


trainFile = customPath + "/val.lists"
file = open(trainFile, 'w')


counter = 0
customPath = 'Your/abs/path/' 
for eachFile in onlyfiles:
  if "jpg" in eachFile:
      counter+=1
      file.write(customPath + eachFile + "\n")


print(counter)

hi! i am a newbie, even i am getting the same error and i use the below code for generating train and test files could you help me by making changes to below code:

import os
import random

imgspath = 'C:/yolo_v4/yolo_v4_mask_detection/darknet/build/darknet/x64/data/obj'
path = 'data/obj/'


images = []
for i in os.listdir(imgspath):
    temp = path+i
    images.append(temp)
# train and test split... adjust it if necessary
trainlen = round(len(images)*.80)
testlen = round(len(images)*.20)
#print('total, train, test dataset size -',trainlen+testlen,trainlen,testlen)
random.shuffle(images)
test = images[:testlen]
train = images[testlen:]

with open('train.txt', 'w') as f:
    for item in train:
        f.write("%s\n" % item)
with open('test.txt', 'w') as f:
    for item in test:
        f.write("%s\n" % item)

[ganapathi12]

@sharoseali

[Shovankumarp]

I have the same problem couldn't solve yet. Could anyone help me please.

[MysticCoss]

I have the same problem here. All i did is DOUBLE CHECK the path or the images

[joseporiolayats]

Hi I had the same problem and I solved it by cloning darknet into a different directory (and compiling, and executing). I'm using c:\darknet as my base directory and now everything works. In my case I suspect it was related to my directories having spaces in the names.

Also I rechecked that all directories were marked with a \ and not \ or /.

I hope it may help.

[giangnd1808]

because it's no data