Requirents-For-Darknet

Label Image:

If your dataset have been collected, then we need to label it for object detection. For label the image, we need to download the tool. I prefer labelImg by tzutalian which is python scripted and it is much easier to use. To get this, we need to clone their repository. link: www.github.com/tzutalin/labelImg All the requirements to run this tool is mentioned there. Apart from this, i made a video how to use it. you can refer it also. link: https://www.youtube.com/watch?v=unxKwZ9_zCc

After Labeling the image we need to download darknet repository.

Clone this : https://github.com/AlexeyAB/darknet

We also need to download pre-trained weights for training purpose.

to download this: Link: https://pjreddie.com/media/files/darknet53.conv.74 (153 MB)

After downloading darknet and pretrained weights:

we need to extract darknet and paste "darknet53.convo.74.weights" in darknet folder. And we have to make it.

After make, we need to modify yolov3-custom, obj.names, obj.data, coco.names, train.txt, test.txt. "data/obj" folder will contain our images and labels.

generate_train.py will generate train.txt file

open cmd on darknet folder, and run this file by typing "python generate_train.py" in cmd. Then u will find a text file named as train.txt in data folder.

In case, if generated train.txt file is empty(0 bytes) then, there is problem in "generate_train.py" of line "if filename.endswith(".jpg"):" Line: "if filename.endswith(".jpg"):" here replace the extension of ur dataset image's extenion. For exapmle, try with .JPG or .JPEG or .png