Open liangxi627 opened 6 years ago
Small object detection is a challenging task for any detection algorithm. In my view, to improve the performance of small object detection is mainly to improve the quality of the feature map used to detect small objects. My first suggestion is to use at least a 38x38 feature map to detect small objects instead of 19x19 feature map used in the original paper. In addition, increasing the resolution of the input image is definitely a big contribution to the performance of the small object detection. Our latest model is specifically designed to be more efficient for high-resolution input. It achieves a pretty good accuracy on the drone image dataset. This model may be open sourced early next year.
Many thanks! Looking forward to your latest model.
Small object detection is a challenging task for any detection algorithm. In my view, to improve the performance of small object detection is mainly to improve the quality of the feature map used to detect small objects. My first suggestion is to use at least a 38x38 feature map to detect small objects instead of 19x19 feature map used in the original paper. In addition, increasing the resolution of the input image is definitely a big contribution to the performance of the small object detection. Our latest model is specifically designed to be more efficient for high-resolution input. It achieves a pretty good accuracy on the drone image dataset. This model may be open sourced early next year.
How to change the feature map from 19x19 to 38x38?
Small object detection is a challenging task for any detection algorithm. In my view, to improve the performance of small object detection is mainly to improve the quality of the feature map used to detect small objects. My first suggestion is to use at least a 38x38 feature map to detect small objects instead of 19x19 feature map used in the original paper. In addition, increasing the resolution of the input image is definitely a big contribution to the performance of the small object detection. Our latest model is specifically designed to be more efficient for high-resolution input. It achieves a pretty good accuracy on the drone image dataset. This model may be open sourced early next year.
Is it open sourced ye?
Small object detection is a challenging task for any detection algorithm. In my view, to improve the performance of small object detection is mainly to improve the quality of the feature map used to detect small objects. My first suggestion is to use at least a 38x38 feature map to detect small objects instead of 19x19 feature map used in the original paper. In addition, increasing the resolution of the input image is definitely a big contribution to the performance of the small object detection. Our latest model is specifically designed to be more efficient for high-resolution input. It achieves a pretty good accuracy on the drone image dataset. This model may be open sourced early next year.
@Robert-JunWang Any update about the model? will be so thankful, if you can provide it
@Robert-JunWang Hello, thanks for your great work! I want to use this network to detect the small objects in a drone dataset. The images are about 1400px1080px, and the objects are only 50px50px. The training mbox_loss maintains at 4 or 5 after 20000 iterations, the detection accuracy is only 30%. Is there any suggestions to modify the network or is there any tricks during training phase? Please help me, thank you~