search

lucasjinreal / yolov7_d2

🔥🔥🔥🔥 (Earlier YOLOv7 not official one) YOLO with Transformers and Instance Segmentation, with TensorRT acceleration! 🔥🔥🔥
GNU General Public License v3.0
3.13k stars 483 forks source link

What is the reason why Swin-T only outputs features 1,2,3 and not 0? #183

Open sarmientoj24 opened 2 years ago

sarmientoj24 commented 2 years ago

I saw from papers and architectures that YOLO only takes 3 feature maps (YOLOv4, 5, 6). Since Swin-T outputs four feature maps, why is that only three features are used here? IS there a way to use all of them or are they necessary?

lucasjinreal commented 2 years ago

you can specific any output features

sarmientoj24 commented 2 years ago

Does that mean I can specify four output features? it seems like YOLO only takes three. is there a better way to use all features with YOLO without loss of information?

sarmientoj24 commented 2 years ago

Was specifically referring to how to accomodate 4 features for YOLOv6 since original YOLOv6 only takes three feature maps