Skip to content

Latest commit

 

History

History
51 lines (40 loc) · 3.17 KB

MODELS.md

File metadata and controls

51 lines (40 loc) · 3.17 KB

Scene Parsing

Accurate Models

Method Backbone ADE20K
(mIoU)
Cityscapes
(mIoU)
COCO-Stuff
(mIoU)
Params
(M)
GFLOPs
(512x512)
GFLOPs
(1024x1024)
Weights
SegFormer MiT-B1 42.2 78.5 40.2 14 16 244 ade
MiT-B2 46.5 81.0 44.6 28 62 717 ade
MiT-B3 49.4 81.7 45.5 47 79 963 ade
Lawin MiT-B1 42.1 79.0 40.5 14 13 218 -
MiT-B2 47.8 81.7 45.2 30 45 563 -
MiT-B3 50.3 82.5 46.6 50 62 809 -
  • mIoU results are with a single scale from official papers.
  • ADE20K image size = 512x512
  • Cityscapes image size = 1024x1024
  • COCO-Stuff image size = 512x512

Real-time Models

Method Backbone CityScapes-val
(mIoU)
CamVid
(mIoU)
Params (M) GFLOPs
(1024x2048)
Weights
BiSeNetv1 ResNet-18 74.8 68.7 14 49 -
BiSeNetv2 - 73.4 72.4 18 21 -
SFNet ResNetD-18 79.0 - 13 - -
FCHarDNet HarDNet-70 77.7 - 4 35 -
DDRNet DDRNet-23slim 77.8 74.7 6 36 cityscapes
  • mIoU results are with a single scale from official papers.
  • Cityscapes image size = 1024x2048 (except BiSeNetv1 & 2 which uses 512x1024)
  • CamVid image size = 960x720

Face Parsing

Method Backbone HELEN-val
(mIoU)
Params
(M)
GFLOPs
(512x512)
FPS
(GTX1660ti)
Weights
BiSeNetv1 MobileNetV2-1.0 58.22 5 5 160 pt
BiSeNetv1 ResNet-18 58.50 14 13 263 pt
BiSeNetv2 - 58.58 18 15 195 pt
FCHarDNet HarDNet-70 59.38 4 4 130 pt
DDRNet DDRNet-23slim 61.11 6 5 180 pt|tflite(fp32)|tflite(fp16)|tflite(int8)
SegFormer MiT-B0 59.31 4 8 75 pt
SFNet ResNetD-18 61.00 14 31 56 pt