π VFNet: IMHO, is the best anchor-free single-stage model, and it's not under the radar.
VariFocalNet: An IoU-aware Dense Object Detector
π§ Background:
π Accurately ranking candidate detections is crucial for dense object detectors to achieve high performance.
...
VariFocalNet: An IoU-aware Dense Object Detector
π§ Background:
π Accurately ranking candidate detections is crucial for dense object detectors to achieve high performance.
...
π Prior work uses the classification score or a combination of classification and predicted localization scores (centerness) to rank candidates.
π Those 2 scores are still not optimal.
π§ Novelty:
π VFNet proposes to learn an IoU-Aware Classification Score (IACS)
π Those 2 scores are still not optimal.
π§ Novelty:
π VFNet proposes to learn an IoU-Aware Classification Score (IACS)
πIACS is used as a joint representation of object presence confidence and localization accuracy using IoU
π VFNet introduces the VariFocal Loss
π The VariFocal Loss down-weights only negative examples for addressing the class imbalance problem during training.
π VFNet introduces the VariFocal Loss
π The VariFocal Loss down-weights only negative examples for addressing the class imbalance problem during training.
π The VariFocal Loss up-weights high-quality positive examples for generating prime detections
π§ VFNet Architecture:
Pushpin VFNet is based on the FCOS+ATSS with the centerness branch removed
π It has three new components:
β¨ The VariFocal Loss,
β¨ The star-shaped bounding box feature representation
β¨ The bounding box refinement
Pushpin VFNet is based on the FCOS+ATSS with the centerness branch removed
π It has three new components:
β¨ The VariFocal Loss,
β¨ The star-shaped bounding box feature representation
β¨ The bounding box refinement
π VFNet also uses GIoU Loss for both bounding boxes branches
π VariFocal Loss consistently improved RetinaNet, FoveaBox and ATSS by 0.9 AP, and by 1.4 AP for RepPoints
π IceVision fully supports the VFNet model!
- IceVision Repo: github.com/airctic/IceVisβ¦
π VariFocal Loss consistently improved RetinaNet, FoveaBox and ATSS by 0.9 AP, and by 1.4 AP for RepPoints
π IceVision fully supports the VFNet model!
- IceVision Repo: github.com/airctic/IceVisβ¦
If you found this thread valuable,
Follow me for more threads on advanced object detection techniques used in real-world applications β
@ai_fast_track
Follow me for more threads on advanced object detection techniques used in real-world applications β
@ai_fast_track
β’ β’ β’
Missing some Tweet in this thread? You can try to
force a refresh
γ
