Monocular Depth Estimation on KITTI 80m maximum depth (Eigen)
0.052Abs RelBinsFormer
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| BinsFormerBackbone=Swin-Large, pre-trained=ImageNet-22K2022.04 | 0.052 | 0.151 | 2.098 | 0.079 | 97.4 | 99.7 | 99.9 | |
| BinsFormerBackbone=Swin-Base, pre-trained=ImageNet-22K2022.04 | 0.053 | 0.156 | 2.141 | 0.08 | 97.4 | 99.7 | 99.9 | |
| BinsFormerBackbone=Swin-Base2022.04 | 0.056 | 0.172 | 2.248 | 0.085 | 97 | 99.6 | 99.9 | |
| AdaBinsBackbone=EfficientNet-B5 + mini-ViT2022.04 | 0.058 | 0.19 | 2.36 | 0.088 | 96.4 | 99.5 | 99.9 | |
| BinsFormerBackbone=Swin-Tiny2022.04 | 0.058 | 0.183 | 2.286 | 0.088 | 96.8 | 99.5 | 99.9 | |
| BTSBackbone=DenseNet-1612022.04 | 0.059 | 0.245 | 2.756 | 0.096 | 95.6 | 99.3 | 99.8 | |
| BinsFormerBackbone=ResNet-502022.04 | 0.061 | 0.208 | 2.426 | 0.093 | 96.2 | 99.4 | 99.9 | |
| DPTBackbone=ResNet-50 + ViT-B, pre-trained=ImageNet-22K2022.04 | 0.062 | — | 2.573 | 0.092 | 95.9 | 99.5 | 99.9 | |
| PGA-NetBackbone=ResNet-502022.04 | 0.063 | 0.267 | 2.634 | 0.101 | 95.2 | 99.2 | 99.8 | |
| TransDepthBackbone=ResNet-50 + ViT-B, auxiliary depth estimation dataset=true2022.04 | 0.064 | 0.252 | 2.755 | 0.098 | 95.6 | 99.4 | 99.9 | |
| BinsFormerBackbone=ResNet-182022.04 | 0.065 | 0.23 | 2.574 | 0.099 | 95.4 | 99.4 | 99.9 | |
| DORN et al.Backbone=ResNet-1012022.04 | 0.072 | 0.307 | 2.727 | 0.12 | 93.2 | 98.4 | 99.4 | |
| Yin et al.Backbone=ResNext-1012022.04 | 0.072 | — | 3.258 | 0.117 | 93.8 | 99 | 99.8 | |
| Gan et al.Backbone=ResNet-1012022.04 | 0.098 | 0.666 | 3.933 | 0.173 | 89 | 96.4 | 98.5 | |
| HR-DepthSupervision=MS, Resolution=1024 x 320, Dataset=K2020.12 | 0.101 | 0.716 | 4.395 | 0.179 | 89.9 | 96.6 | 98.3 | |
| Lite-HR-DepthSupervision=T, Resolution=1280 x 384, #Para=3.1M2020.12 | 0.104 | 0.697 | 4.349 | 0.176 | 89.3 | 96.7 | 98.5 | |
| HR-DepthSupervision=MS, Resolution=640 × 192, Dataset=CS + K2020.12 | 0.104 | 0.786 | 4.544 | 0.182 | 89.3 | 96.4 | 98.3 | |
| HR-DepthSupervision=M, Resolution=1280 × 384, Dataset=K2020.12 | 0.104 | 0.727 | 4.41 | 0.179 | 89.4 | 96.6 | 98.4 | |
| Lite-HR-DepthSupervision=T, Resolution=1024 x 320, #Para=3.1M2020.12 | 0.105 | 0.704 | 4.374 | 0.178 | 89.1 | 96.7 | 98.5 | |
| Monodepth2Supervision=MS, Resolution=640 × 192, Dataset=K2020.12 | 0.106 | 0.818 | 4.75 | 0.196 | 87.4 | 95.7 | 97.9 | |
| HR-DepthSupervision=M, Resolution=1024 x 320, Dataset=K2020.12 | 0.106 | 0.755 | 4.472 | 0.181 | 89.2 | 96.6 | 98.4 | |
| Monodepth2Supervision=MS, Resolution=1024 × 320, Dataset=K2020.12 | 0.106 | 0.818 | 4.75 | 0.196 | 87.4 | 95.7 | 97.9 | |
| Johnston et al.Backbone=ResNet-1012022.04 | 0.106 | 0.861 | 4.699 | 0.185 | 88.9 | 96.2 | 98.2 | |
| HR-DepthSupervision=MS, Resolution=640 × 192, Dataset=K2020.12 | 0.107 | 0.785 | 4.612 | 0.185 | 88.7 | 96.2 | 98.2 | |
| PackNet-SfMSupervision=M, Resolution=1280 × 384, Dataset=K2020.12 | 0.107 | 0.802 | 4.538 | 0.186 | 88.9 | 96.2 | 98.1 | |
| HR-DepthSupervision=M, Resolution=640 × 192, Dataset=CS + K2020.12 | 0.108 | 0.955 | 4.8 | 0.19 | 88.7 | 96.1 | 98.1 | |
| struct2depth (M+R)Motion?=M, Cap=80m, Refinement=true2018.11 | 0.1087 | 0.825 | 4.7503 | 0.1866 | 87.38 | 95.77 | 98.25 | |
| HR-DepthSupervision=M, Resolution=640 × 192, Dataset=K2020.12 | 0.109 | 0.792 | 4.632 | 0.185 | 88.4 | 96.2 | 98.3 | |
| 3NetSupervision=Stereo, Train set=CS+E, Backbone=ResNet50, Post-processing=true2018.08 | 0.111 | 0.849 | 4.822 | 0.202 | 86.5 | 95.2 | 97.8 | |
| Lite-HR-DepthSupervision=M, Resolution=1024 x 320, #Para=3.1M2020.12 | 0.111 | 0.799 | 4.612 | 0.184 | 87.8 | 96.3 | 98.4 | |
| PackNet-SfMSupervision=M, Resolution=640 × 192, Dataset=K2020.12 | 0.111 | 0.785 | 4.601 | 0.189 | 87.8 | 96 | 98.2 | |
| Lite-HR-DepthSupervision=T, Resolution=640 x 192, #Para=3.1M2020.12 | 0.112 | 0.767 | 4.557 | 0.184 | 87.6 | 96.3 | 98.4 | |
| 3NetSupervision=Stereo, Train set=CS+E, Backbone=ResNet502018.08 | 0.113 | 0.885 | 4.898 | 0.204 | 86.2 | 95 | 97.7 | |
| Kuznietsov et al.Supervised=Semi, Training Dataset=K, Cap=80m2019.04 | 0.113 | 0.741 | 4.621 | 0.189 | 86.2 | 96 | 98.6 | |
| Godard et al.Supervision=Stereo, Train set=CS+E, Backbone=ResNet50, Post-processing=true2018.08 | 0.114 | 0.898 | 4.935 | 0.206 | 86.1 | 94.9 | 97.6 | |
| Godard et al.Backbone=ResNet-502022.04 | 0.114 | 0.898 | 4.935 | 0.206 | 86.1 | 94.9 | 97.6 | |
| Monodepth2Supervision=M, Resolution=640 x 192, #Para=14.84M2020.12 | 0.115 | 0.903 | 4.863 | 0.193 | 87.7 | 95.9 | 98.1 | |
| Monodepth2Supervision=M, Resolution=1024 x 320, #Para=14.84M2020.12 | 0.115 | 0.882 | 4.701 | 0.19 | 87.9 | 96.1 | 98.2 | |
| Monodepth2Supervision=M, Resolution=640 × 192, Dataset=K2020.12 | 0.115 | 0.903 | 4.863 | 0.193 | 87.7 | 95.9 | 98.1 | |
| Monodepth2Supervision=M, Resolution=1024 × 320, Dataset=K2020.12 | 0.115 | 0.882 | 4.701 | 0.19 | 87.9 | 96.1 | 98.2 | |
| SharinGANSupervised=No, Training Dataset=K+S, Cap=80m2020.06 | 0.116 | 0.939 | 5.068 | 0.203 | 85 | 94.8 | 97.8 | |
| Lite-HR-DepthSupervision=M, Resolution=640 x 192, #Para=3.1M2020.12 | 0.116 | 0.845 | 4.841 | 0.19 | 86.6 | 95.7 | 98.2 | |
| 3NetSupervision=Stereo, Train set=CS+E, Backbone=VGG2018.08 | 0.117 | 0.905 | 4.982 | 0.21 | 85.6 | 94.8 | 97.6 | |
| Godard et al.Supervision=Stereo, Train set=CS+E, Backbone=ResNet502018.08 | 0.121 | 1.037 | 5.212 | 0.216 | 85.4 | 94.4 | 97.3 | |
| Zhou et al.Supervision=M, Resolution=1248 × 384, Dataset=K2020.12 | 0.121 | 0.837 | 4.945 | 0.197 | 85.3 | 95.5 | 98.2 | |
| struct2depth (R)Cap=80m, Refinement=true2018.11 | 0.1231 | 1.4367 | 5.3099 | 0.2043 | 87.05 | 95.14 | 97.65 | |
| Godard et al.Supervision=Stereo, Train set=CS+E, Backbone=VGG2018.08 | 0.124 | 1.076 | 5.311 | 0.219 | 84.7 | 94.2 | 97.3 | |
| 3NetSupervision=Stereo, Train set=E, Backbone=ResNet50, Post-processing=true2018.08 | 0.126 | 0.961 | 5.205 | 0.22 | 83.5 | 94.1 | 97.4 | |
| Godard et al.Supervision=Stereo, Train set=E, Backbone=ResNet50, Post-processing=true2018.08 | 0.128 | 1.038 | 5.355 | 0.223 | 83.3 | 93.9 | 97.2 | |
| 3NetSupervision=Stereo, Train set=E, Backbone=ResNet502018.08 | 0.129 | 0.996 | 5.281 | 0.223 | 83.1 | 93.9 | 97.4 | |
| Yang (2018a) [Motion]Motion?=M, Cap=80m2018.11 | 0.131 | 1.254 | 6.117 | 0.22 | 82.6 | 93.1 | 97.3 | |
| Godard et al.Supervision=Stereo, Train set=E, Backbone=ResNet502018.08 | 0.133 | 1.142 | 5.533 | 0.23 | 83 | 93.6 | 97 | |
| GodardCap=80m2018.11 | 0.133 | 1.158 | 5.37 | 0.208 | 84.1 | 94.9 | 97.8 | |
| Yang (2018a) [Static]Cap=80m2018.11 | 0.137 | 1.326 | 6.232 | 0.224 | 80.6 | 92.7 | 97.3 | |
| CC (ours)Data=cs+k, Supervision=Unsupervised2018.05 | 0.139 | 1.032 | 5.199 | 0.213 | 82.7 | 94.3 | 97.7 | |
| CC (ours)Data=k, Supervision=Unsupervised2018.05 | 0.14 | 1.07 | 5.326 | 0.217 | 82.6 | 94.1 | 97.5 | |
| Struct2DepthSupervision=M, Resolution=416 × 128, Dataset=K2020.12 | 0.141 | 1.026 | 5.291 | 0.215 | 81.6 | 94.5 | 97.9 | |
| struct2depth (M)Motion?=M, Cap=80m2018.11 | 0.1412 | 1.0258 | 5.2905 | 0.2153 | 81.6 | 94.52 | 97.91 | |
| struct2depth (Baseline)Cap=80m2018.11 | 0.1417 | 1.1385 | 5.5205 | 0.2186 | 82.03 | 94.15 | 97.62 | |
| 3NetSupervision=Stereo, Train set=E, Backbone=VGG2018.08 | 0.142 | 1.207 | 5.702 | 0.24 | 80.9 | 92.8 | 96.7 | |
| Zhan et al.Supervision=Stereo+Temp., Train set=E2018.08 | 0.144 | 1.391 | 5.869 | 0.241 | 80.3 | 92.8 | 96.9 | |
| PyD-Net (200)Supervision=Stereo, Train set=CS+E2018.08 | 0.146 | 1.291 | 5.907 | 0.245 | 80.1 | 92.6 | 96.7 | |
| DF-NetData=cs+k, Supervision=Unsupervised2018.05 | 0.146 | 1.182 | 5.215 | 0.213 | 81.8 | 94.3 | 97.8 | |
| PyD-NetSupervision=M, Resolution=512 x 256, #Para=1.9M2020.12 | 0.146 | 1.291 | 5.907 | 0.245 | 80.1 | 92.6 | 96.7 | |
| Godard et al.Supervision=Stereo, Train set=E, Backbone=VGG2018.08 | 0.148 | 1.344 | 5.927 | 0.247 | 80.3 | 92.2 | 96.4 | |
| Wang et al.Supervision=Temporal, Train set=CS+E2018.08 | 0.148 | 1.187 | 5.496 | 0.226 | 81.2 | 93.8 | 97.5 | |
| Godard et al.Supervised=No, Training Dataset=K, Cap=80m2019.04 | 0.148 | 1.344 | 5.927 | 0.247 | 80.3 | 92.2 | 96.4 | |
| GASDASupervised=No, Training Dataset=K+S(DA), Cap=80m2019.04 | 0.149 | 1.003 | 4.995 | 0.227 | 82.4 | 94.1 | 97.3 | |
| GASDASupervised=No, Training Dataset=K+S, Cap=80m2020.06 | 0.149 | 1.003 | 4.995 | 0.227 | 82.4 | 94.1 | 97.3 | |
| DF-NetData=k, Supervision=Unsupervised2018.05 | 0.15 | 1.124 | 5.507 | 0.223 | 80.6 | 93.3 | 97.3 | |
| Wang et al.Supervision=Temporal, Train set=E2018.08 | 0.151 | 1.257 | 5.583 | 0.228 | 81 | 93.6 | 97.4 | |
| DDVOCap=80m2018.11 | 0.151 | 1.257 | 5.583 | 0.228 | 81 | 93.6 | 97.4 | |
| PyD-Net (200)Supervision=Stereo, Train set=E2018.08 | 0.153 | 1.363 | 6.03 | 0.252 | 78.9 | 91.8 | 96.3 | |
| GeoNet ResNet50Supervision=Temporal, Train set=CS+E, Backbone=ResNet502018.08 | 0.153 | 1.328 | 5.737 | 0.232 | 80.2 | 93.4 | 97.2 | |
| Geonet-ResnetData=cs+k, Supervision=Unsupervised, Backbone=ResNet2018.05 | 0.153 | 1.328 | 5.737 | 0.232 | 80.2 | 93.4 | 97.2 | |
| Godard et al.Data=k, Supervision=Unsupervised2018.05 | 0.154 | 1.218 | 5.699 | 0.231 | 79.8 | 93.2 | 97.3 | |
| GeoNet ResNet50Supervision=Temporal, Train set=E, Backbone=ResNet502018.08 | 0.155 | 1.296 | 5.857 | 0.233 | 79.3 | 93.1 | 97.3 | |
| GeoNetMotion?=M, Cap=80m2018.11 | 0.155 | 1.296 | 5.857 | 0.233 | 79.3 | 93.1 | 97.3 | |
| Geonet-ResnetData=k, Supervision=Unsupervised, Backbone=ResNet2018.05 | 0.155 | 1.296 | 5.857 | 0.233 | 79.3 | 93.1 | 97.3 | |
| All real (baseline2)Supervised=No, Training Dataset=K, Cap=80m2019.04 | 0.158 | 1.151 | 5.285 | 0.238 | 81.1 | 93.4 | 97 | |
| All real (baseline)Supervised=No, Training Dataset=K, Cap=80m2020.06 | 0.158 | 1.151 | 5.285 | 0.238 | 81.1 | 93.4 | 97 | |
| All realSupervised=No, Dataset=K2024.05 | 0.158 | 1.151 | 5.285 | 0.238 | 81.1 | 93.4 | 97 | |
| Mahjourian et al.Supervision=Temporal, Train set=CS+E2018.08 | 0.159 | 1.231 | 5.912 | 0.243 | 78.4 | 92.3 | 97 | |
| Mahjourian et al.Data=cs+k, Supervision=Unsupervised2018.05 | 0.159 | 1.231 | 5.912 | 0.243 | 78.4 | 92.3 | 97 | |
| Vid2DepthSupervision=M, Resolution=416 x 128, Dataset=CS + K2020.12 | 0.159 | 1.231 | 5.912 | 0.243 | 78.4 | 92.3 | 97 | |
| LEGOMotion?=M, Cap=80m2018.11 | 0.162 | 1.352 | 6.276 | 0.252 | 78.3 | 92.1 | 96.9 | |
| Mahjourian et al.Supervision=Temporal, Train set=E2018.08 | 0.163 | 1.24 | 6.22 | 0.25 | 76.2 | 91.6 | 96.8 | |
| Vid2DepthCap=80m2018.11 | 0.163 | 1.24 | 6.22 | 0.25 | 76.2 | 91.6 | 96.8 | |
| Mahjourian et al.Data=k, Supervision=Unsupervised2018.05 | 0.163 | 1.24 | 6.22 | 0.25 | 76.2 | 91.6 | 96.8 | |
| GeoNetSupervision=Temporal, Train set=E2018.08 | 0.164 | 1.303 | 6.09 | 0.247 | 76.5 | 91.9 | 96.8 | |
| Geonet-VGGData=k, Supervision=Unsupervised, Backbone=VGG2018.05 | 0.164 | 1.303 | 6.09 | 0.247 | 76.5 | 91.9 | 96.8 | |
| Kundu et al.Supervised=Semi, Training Dataset=K+S(DA), Cap=80m2019.04 | 0.167 | 1.257 | 5.578 | 0.237 | 77.1 | 92.2 | 97.1 | |
| SemiSupMDESupervised=No, Dataset=K+S(DA)2024.05 | 0.168 | 1.347 | 6.002 | 0.254 | 76.9 | 90.7 | 96.1 | |
| Yen et al.Supervised=No, Dataset=K+S(DA)2024.05 | 0.169 | 1.371 | 6.037 | 0.256 | 75.9 | 90.4 | 96.1 | |
| Yang (2017)Cap=80m2018.11 | 0.182 | 1.481 | 6.501 | 0.267 | 72.5 | 90.6 | 96.3 | |
| Zheng et al.Supervised=No, Dataset=K+S(DA)2024.05 | 0.182 | 1.611 | 6.216 | 0.265 | 74.9 | 89.8 | 95.9 | |
| Zhou et al. (updated [40])Supervision=Temporal, Train set=E2018.08 | 0.183 | 1.595 | 6.709 | 0.27 | 73.4 | 90.2 | 95.9 | |
| Zhou et al.*Data=k, Supervision=Unsupervised, Source=github page2018.05 | 0.183 | 1.595 | 6.709 | 0.27 | 73.4 | 90.2 | 95.9 | |
| Zhou et al.Supervision=Temporal, Train set=CS+E2018.08 | 0.198 | 1.836 | 6.565 | 0.275 | 71.8 | 90.1 | 96 | |
| Zhou et al.Data=cs+k, Supervision=Unsupervised2018.05 | 0.198 | 1.836 | 6.565 | 0.275 | 71.8 | 90.1 | 96 |