diff --git a/slides/05_conv_nets_2/index.html b/slides/05_conv_nets_2/index.html index 7fdd3d0..adfc0f2 100644 --- a/slides/05_conv_nets_2/index.html +++ b/slides/05_conv_nets_2/index.html @@ -326,7 +326,7 @@ Region proposal input is a fully convolutional network: shares weights across spatial dimensions --- -## State-of-the-art +## Measuring performance .center[ @@ -342,25 +342,6 @@ - AP @0.5 for class "cat": average precision for the class, where $IoU(box^{pred}, box^{true}) > 0.5$ ---- -## State-of-the-art - fast methods - -.center[ - -] - -Measures: **mAP** and Frames per Second **FPS** - -.footnote.small[ -Zeming Li et al. Light-Head R-CNN: In Defense of Two-Stage Object Detector 2017 -] - --- - -- Mask RCNN, light-head R-CNN for best accuracy -- Yolo, SSD, Light-Head R-CNN for fast inference - - --- ## State-of-the-art