tiny-yolo not giving results after fine tuning

[sharoseali]

@AlexeyAB writing stopbackward=1 in cfg file and then fine tuning isn't giving results as yolov2 23 layer architecture. mAP is 0.0% before and after fine tuning yolov2-tiny-voc.cfg and corresponding weights

this result in a pic is after first 100 iterations......... After 2400 iteration the avg loss donot decrease below 0.2 and mAP is 0.00% ..................extremely un-understandable
Alexey. after run this command
darknet partial cfg/yolov2-tiny-voc.cfg yolov2-tiny-voc.weights yolov2-tiny.conv.13 13
the weights are saved as yolov2-tiny.conv.13 and when i tested this weight file for detection it gives me this result

if this command only save the weight file than why results are so bad??

[AlexeyAB]

after run this command
darknet partial cfg/yolov2-tiny-voc.cfg yolov2-tiny-voc.weights yolov2-tiny.conv.13 13
the weights are saved as yolov2-tiny.conv.13 and when i tested this weight file for detection it gives me this result

This command creates weights file yolov2-tiny.conv.13 that is suitable only for Training, not for Detection:
darknet partial cfg/yolov2-tiny-voc.cfg yolov2-tiny-voc.weights yolov2-tiny.conv.13 13

---

You must follow exactly this sequence:

1. Set stopbackward=1
2. do darknet partial to get yolov2-tiny.conv.13 file
3. Train your model yolo-obj.weights
4. use yolo-obj.weights for Detection

Read more: https://github.com/AlexeyAB/darknet#how-to-train-tiny-yolo-to-detect-your-custom-objects
and: https://github.com/AlexeyAB/darknet#how-to-improve-object-detection

[sharoseali]

@AlexeyAB Thanks for replying... i set stopbackward=1 in my new cfg file to train with 4 classes.
Now i will try again by setting stopbackward=1 in tiny-yolo-voc.cfg before partial command. Let see what happens??

[sharoseali]

@alexey not getting the results even then...... i don't understand what's wrong with tiny- yolo-voc.cfg..
Avg loss decreases in several minutes and struck around 2.0.... mAP is 0.0%...