Swin-V2 as backbone for One-Stage Detector has low precision and recall for area==medium

[sarmientoj24]

I was using Swin-Transformerv2 for object detection as a backbone to a one stage detector. While the metrics are competitive, I am getting very low scores on [email protected]:0.95 because the Precision and Recall on area=medium is ridiculously low. Any insights on that? Does using Swin-Transformerv2 have a direct effect on this?

Swin-YOLO

 Average Precision  (AP) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.589
 Average Precision  (AP) @[ IoU=0.50      | area=   all | maxDets=100 ] = 0.747
 Average Precision  (AP) @[ IoU=0.75      | area=   all | maxDets=100 ] = 0.703
 Average Precision  (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = -1.000
 Average Precision  (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.125  <---------low
 Average Precision  (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.589
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=  1 ] = 0.736
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets= 10 ] = 0.809
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.809
 Average Recall     (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = -1.000
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.500 <---------low
 Average Recall     (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.809


YOLO
 Average Precision  (AP) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.660
 Average Precision  (AP) @[ IoU=0.50      | area=   all | maxDets=100 ] = 0.793
 Average Precision  (AP) @[ IoU=0.75      | area=   all | maxDets=100 ] = 0.759
 Average Precision  (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = -1.000
 Average Precision  (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.700 <---------high
 Average Precision  (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.661
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=  1 ] = 0.795
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets= 10 ] = 0.842
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.842
 Average Recall     (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = -1.000
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.700 <---------high
 Average Recall     (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.842
 

[Bailey-24]

Hi, how could you create your swinv2, and how do you extract the feature from swinv2, such as my output size is torch.size([1, 1024]), but I want the size is torch.size([1, 1024, 20, 20]) please show me the code. Thanks in advanced.