Fine-tuned models

[WeiquanWa]

The result looks like cool. I'm pretty wonder if yolov4-tiny can produce such a good detection accuracy even after fine-tuned with a custom dataset. It would be cool if you would upload your fine-tuned YoloV4-tiny model.

[jaydubal]

Hi, can you please provide your "yolov4-tiny.weights", "yolov4-tiny.names" & "yolov4-tiny.cfg"

[sriks6711]

The result looks like cool. I'm pretty wonder if yolov4-tiny can produce such a good detection accuracy even after fine-tuned with a custom dataset. It would be cool if you would upload your fine-tuned YoloV4-tiny model.

The model (or rather the .weights file in Darknet parlance) is available in the backup folder.

In particular, the weights file used in the command [1] is here -
https://github.com/achen353/Taiwanese-Traffic-Object-Detection/blob/master/backup/yolov4-tiny-obj_best.weights

Hope it helps.

[1] $./darknet detector demo data/obj.data cfg/yolov4-tiny-obj.cfg yolov4-tiny-obj_best.weights sample_videos/test.mp4

[sriks6711]

Hi, can you please provide your "yolov4-tiny.weights", "yolov4-tiny.names" & "yolov4-tiny.cfg"

@jaydubal - It would be best if you logged this as a separate issue to get the attention of @achen353

[jaydubal]

Edit: Nevermind, I saw you opened an issue for this.
@sriks6711 thanks. Can you please also guide which .cfg and .names to use? because using the mentioned weights, it is not working as expected. It will detect each and every object of .names in the provided image.

[sriks6711]

Edit: Nevermind, I saw you opened an issue for this. @sriks6711 thanks. Can you please also guide which .cfg and .names to use? because using the mentioned weights, it is not working as expected. It will detect each and every object of .names in the provided image.

We got somewhere and will share them in a separate ticket

[sriks6711]

Edit: Nevermind, I saw you opened an issue for this. @sriks6711 thanks. Can you please also guide which .cfg and .names to use? because using the mentioned weights, it is not working as expected. It will detect each and every object of .names in the provided image.

@jaydubal

There should be 4 classes/labels in the names file you use -
vehicle
scooter
pedestrian
bicycle

The reference is https://github.com/achen353/Taiwanese-Traffic-Object-Detection#dataset
"... dataset consists of 89002 images of size 1920*1080. There are 4 annotated classes: vehicle, scooter, pedestrian, bicycle ..."