search

WongKinYiu / yolor

implementation of paper - You Only Learn One Representation: Unified Network for Multiple Tasks (https://arxiv.org/abs/2105.04206)
GNU General Public License v3.0
1.98k stars 524 forks source link

yoloR not supporting background images?? Too many False predictions compaired to yolov4x-mish-darknet #104

Open akashAD98 opened 2 years ago

akashAD98 commented 2 years ago

I have trained the model using yolov4x-mish using darknet & added background images to avoid false prediction. Used same dataset to train yoloR. I tried yolor-p6 with 416 image resolution.

Results of yoloR good,Also it's detecting small objects well compaired to yolov4x-mish model. But im getting too many false predictions, Even after adding background images.

So yoloR not supporting background images?? If its supporting why there are too many FP results? How can we solve the FP problem? @WongKinYiu

I have tested on 50% confidence for both Yolo & yoloR model.

left one is YoloR & Right one is yolov4x-mish-

fpyolor

yolor_extradetect

FP_compair_yolo_yoloR_2

FP_compair_yolo_yoloR_1

Extra/correct detection by yoloR

note-(red box-not detected -yolov4x-mish) yolor-working well for detecting small parts, but there are too many FP too.

yolor_extra_detect

yolor_extradetect

FP_compair_yolo_yoloR_4_extra_detect

akashAD98 commented 2 years ago

@WongKinYiu @AlexeyAB can you please guide me? How should i solve this FP problem? Why yoloR Not supporting Background images? is there any other alternative for this issue? Looking forward for your views on this issue. Thankyou so much