Skip to content

Latest commit

 

History

History
57 lines (51 loc) · 6.59 KB

MODEL_ZOO.md

File metadata and controls

57 lines (51 loc) · 6.59 KB
Raw

Arbitrary-Oriented Object Detection with Circular Smooth Label

Performance（deprecated）

Due to the improvement of the code, the performance of this repo is gradually improving, so the experimental results in this file are for reference only.

Window Function

Model Backbone Training data Val data mAP Model Link Anchor Label Mode Reg. Loss Angle Range lr schd Data Augmentation GPU Image/GPU Configs
CSL ResNet50_v1d 600->800 DOTA1.0 trainval DOTA1.0 test 39.52 - H Pulse smooth L1 90 1x × 1X GeForce RTX 2080 Ti 1 cfgs_res50_dota_v20.py
CSL ResNet50_v1d 600->800 DOTA1.0 trainval DOTA1.0 test 58.86 - H Rectangular smooth L1 90 1x × 1X GeForce RTX 2080 Ti 1 cfgs_res50_dota_v21.py
CSL ResNet50_v1d 600->800 DOTA1.0 trainval DOTA1.0 test 60.15 - H Triangle smooth L1 90 1x × 1X GeForce RTX 2080 Ti 1 cfgs_res50_dota_v22.py
CSL ResNet50_v1d 600->800 DOTA1.0 trainval DOTA1.0 test 63.51 - H Gaussian smooth L1 90 2x × 2X GeForce RTX 2080 Ti 1 cfgs_res50_dota_v18.py
CSL ResNet50_v1d 600->800 DOTA1.0 trainval DOTA1.0 test 42.06 - H Pulse smooth L1 180 2x × 4X GeForce RTX 2080 Ti 1 cfgs_res50_dota_v28.py
CSL ResNet50_v1d 600->800 DOTA1.0 trainval DOTA1.0 test 61.98 - H Rectangular smooth L1 180 2x × 2X GeForce RTX 2080 Ti 1 cfgs_res50_dota_v23.py
CSL ResNet50_v1d 600->800 DOTA1.0 trainval DOTA1.0 test 57.94 - H Triangle smooth L1 180 2x × 4X GeForce RTX 2080 Ti 1 cfgs_res50_dota_v26.py
CSL ResNet50_v1d 600->800 DOTA1.0 trainval DOTA1.0 test 64.50 - H Gaussian smooth L1 180 2x × 2X Quadro RTX 8000 1 cfgs_res50_dota_v27.py
CSL ResNet50_v1d 600->800 DOTA1.0 trainval DOTA1.0 test 65.09 - H Gaussian smooth L1 + atan(theta) 180 2x × 2X Quadro RTX 8000 1 cfgs_res50_dota_v31.py
CSL ResNet50_v1d 600->800 DOTA1.0 trainval DOTA1.0 test 65.44 - H Gaussian smooth L1 + atan(theta) 180 2x × 2X Quadro RTX 8000 1 cfgs_res50_dota_v37.py
CSL ResNet152_v1d MS DOTA1.0 trainval DOTA1.0 test 70.29 model H Gaussian smooth L1 + atan(theta) 180 2x √ 2X Quadro RTX 8000 1 cfgs_res152_dota_v36.py

CSL VS Baseline

Based Method Angle Range Angle Pred. Label Mode BR SV LV SH HA 5-mAP
RetinaNet-H 90 Reg. Five-Param. - 41.15 53.75 48.30 55.92 55.77 50.98
RetinaNet-R 90 Reg. Five-Param. - 32.27 64.64 71.01 68.62 53.52 58.01
RetinaNet-H 180 Reg. Five-Param. - 38.47 54.15 47.89 60.87 53.63 51.00
RetinaNet-H - Reg. Eight-Param. - 43.97 58.50 54.79 65.55 55.65 55.69
RetinaNet-R 90 Cls. Gaussian 35.14 63.21 73.92 69.49 55.53 59.46
RetinaNet-H 90 Cls. Pulse 9.80 28.04 11.42 18.43 23.35 18.21
RetinaNet-H 90 Cls. Rectangular 37.62 54.28 48.97 62.59 50.26 50.74
RetinaNet-H 90 Cls. Triangle 37.25 54.45 44.01 60.03 52.20 49.59
RetinaNet-H 90 Cls. Gaussian 41.03 59.63 52.57 64.56 54.64 54.49
RetinaNet-H 180 Cls. Pulse 13.95 16.79 6.5 16.80 22.48 15.30
RetinaNet-H 180 Cls. Rectangular 36.14 60.80 50.01 65.75 53.17 53.17
RetinaNet-H 180 Cls. Triangle 32.69 47.25 44.39 54.11 41.90 44.07
RetinaNet-H 180 Cls. Gaussian 41.16 63.68 55.44 65.85 55.23 56.21

Radius

Model Backbone Training data Val data mAP Model Link Anchor Label Mode Raduius/Sigma Reg. Loss Angle Range lr schd Data Augmentation GPU Image/GPU Configs
CSL ResNet50_v1d 600->800 DOTA1.0 trainval DOTA1.0 test 63.51 - H Gaussian 4 smooth L1 90 2x × 2X GeForce RTX 2080 Ti 1 cfgs_res50_dota_v18.py
CSL ResNet50_v1d 600->800 DOTA1.0 trainval DOTA1.0 test 65.45 - R Gaussian 4 smooth L1 90 2x × 2X Quadro RTX 8000 1 cfgs_res50_dota_v33.py
CSL ResNet50_v1d 600->800 DOTA1.0 trainval DOTA1.0 test 40.78 - H Gaussian 0.1 smooth L1 180 2x × 2x GeForce RTX 2080 Ti 1 cfgs_res50_dota_v35.py
CSL ResNet50_v1d 600->800 DOTA1.0 trainval DOTA1.0 test 59.23 - H Gaussian 2 smooth L1 180 2x × 2x GeForce RTX 2080 Ti 1 cfgs_res50_dota_v32.py
CSL ResNet50_v1d 600->800 DOTA1.0 trainval DOTA1.0 test 62.12 - H Gaussian 4 smooth L1 180 2x × 4x GeForce RTX 2080 Ti 1 cfgs_res50_dota_v30.py
CSL ResNet50_v1d 600->800 DOTA1.0 trainval DOTA1.0 test 64.50 - H Gaussian 6 smooth L1 180 2x × 2X Quadro RTX 8000 1 cfgs_res50_dota_v27.py
CSL ResNet50_v1d 600->800 DOTA1.0 trainval DOTA1.0 test 63.99 - H Gaussian 8 smooth L1 180 2x × 4x GeForce RTX 2080 Ti 1 cfgs_res50_dota_v29.py
CSL ResNet50_v1d 800->1024 DOTA1.0 trainval DOTA1.0 test 63.68 - H Gaussian 6 smooth L1 180 2x × 2X Quadro RTX 8000 1 cfgs_res50_dota_v25.py

Scene Text Dataset

Backbone Training data Val data Performance (RetinaNet-H) Performance (CSL) GPU Configs
ResNet101_v1d MS ICDAR2015 train ICDAR2015 test 72.12 / 74.90 / 73.49 75.78 / 79.78 / 77.73 2X Quadro RTX 8000 cfgs_res101_icdar2015_v1.py
ResNet101_v1d MS MLT trainval MLT test 2X Quadro RTX 8000
ResNet101_v1d MS MLT trainval + ICDAR2015 train ICDAR2015 test 2X Quadro RTX 8000
ResNet101_v1d MS HRSC2016 train HRSC2016 test 2X Quadro RTX 8000