Depth Prediction on NYU-depth v2 (test)
99.5Accuracy (δ < 1.25)GuideNet
Evaluation results
| Method | Links | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| GuideNetsamples=5002019.08 | 99.5 | 99.9 | 100 | 0.015 | — | 0.101 | — | — | — | |
| DeepLiDARsamples=5002019.08 | 99.3 | 99.9 | 100 | 0.022 | — | 0.115 | — | — | — | |
| CSPNsamples=5002019.08 | 99.2 | 99.9 | 100 | 0.016 | — | 0.117 | — | — | — | |
| UNet+CSPN#Samples=500, Sparsity level=0.72%2018.11 | 99.2 | 99.9 | 100 | 0.016 | — | 0.117 | — | — | — | |
| MS-Net[LF]#Samples=500, Sparsity level=0.72%2018.11 | 99.1 | 99.8 | 100 | 0.018 | — | 0.129 | — | — | — | |
| EncDec-Net[EF]#Samples=500, Sparsity level=0.72%2018.11 | 99.1 | 99.8 | 100 | 0.017 | — | 0.123 | — | — | — | |
| NConv-CNNsamples=5002019.08 | 99 | 99.8 | 100 | 0.018 | — | 0.129 | — | — | — | |
| CSPN#Samples=500, Sparsity level=0.72%2018.11 | 99 | 99.8 | 100 | 0.021 | — | 0.136 | — | — | — | |
| GuideNetsamples=2002019.08 | 98.8 | 99.8 | 100 | 0.024 | — | 0.142 | — | — | — | |
| UNet+SPN#Samples=500, Sparsity level=0.72%2018.11 | 98.8 | 99.8 | 100 | 0.022 | — | 0.144 | — | — | — | |
| SPN#Samples=500, Sparsity level=0.72%2018.11 | 98.5 | 99.7 | 99.9 | 0.027 | — | 0.162 | — | — | — | |
| EncDec-Net[EF]#Samples=200, Sparsity level=0.28%2018.11 | 98.3 | 99.6 | 99.9 | 0.026 | — | 0.171 | — | — | — | |
| NConv-CNNsamples=2002019.08 | 98.2 | 99.6 | 99.9 | 0.027 | — | 0.173 | — | — | — | |
| MS-Net[LF]#Samples=200, Sparsity level=0.28%2018.11 | 97.9 | 99.5 | 99.8 | 0.03 | — | 0.192 | — | — | — | |
| Ma et al.samples=5002019.08 | 97.8 | 99.6 | 99.9 | 0.043 | — | 0.204 | — | — | — | |
| Sparse-to-Dense#Samples=500, Sparsity level=0.72%2018.11 | 97.8 | 99.5 | 99.9 | 0.043 | — | 0.224 | — | — | — | |
| Ma and KaramanPartially known depths=true2018.03 | 97.1 | 99.4 | 99.8 | 0.044 | — | — | — | — | — | |
| Zhang et al.samples=5002019.08 | 97.1 | 99.3 | 99.7 | 0.042 | — | 0.228 | — | — | — | |
| Ma et al.samples=2002019.08 | 97.1 | 99.4 | 99.8 | 0.044 | — | 0.23 | — | — | — | |
| Sparse-to-Dense#Samples=200, Sparsity level=0.28%2018.11 | 97.1 | 99.4 | 99.8 | 0.044 | — | 0.23 | — | — | — | |
| Freeformoptical design=deep optics, aberration/mask type=Optimized freeform lens2019.04 | 93 | 99 | 99.9 | 0.087 | 0.052 | — | — | — | 0.433 | |
| Bilateralsamples=5002019.08 | 92.4 | 97.6 | 98.9 | 0.084 | — | 0.479 | — | — | — | |
| Astigmatismoptical design=deep optics, aberration/mask type=Astigmatism2019.04 | 91.6 | 98.6 | 99.8 | 0.095 | 0.056 | — | — | — | 0.456 | |
| Chromaticoptical design=deep optics, aberration/mask type=Chromatic2019.04 | 91.6 | 98.7 | 99.8 | 0.095 | 0.056 | — | — | — | 0.45 | |
| DenseDepthmedian scaling=true2018.12 | 89.5 | 98 | 99.6 | 0.103 | 0.043 | — | — | — | 0.39 | |
| Defocusoptical design=deep optics, aberration/mask type=Defocus2019.04 | 89.3 | 98.1 | 99.6 | 0.108 | 0.062 | — | — | — | 0.481 | |
| Liao et al.#Samples=2252018.11 | 87.8 | 96.4 | 98.9 | 0.104 | — | 0.442 | — | — | — | |
| Ours (SENet-154)Backbone=SENet-154, Training samples=50K2018.03 | 86.6 | 97.5 | 99.3 | 0.115 | 0.05 | — | — | — | 0.53 | |
| Ours (DenseNet-161)Backbone=DenseNet-161, Training samples=50K2018.03 | 85.5 | 97.2 | 99.3 | 0.123 | 0.053 | — | — | — | 0.544 | |
| DenseDepthscaled=false2018.12 | 84.6 | 97.4 | 99.4 | 0.123 | 0.053 | — | — | — | 0.465 | |
| Ours (ResNet-50)Backbone=ResNet-50, Training samples=50K2018.03 | 84.3 | 96.8 | 99.1 | 0.126 | 0.054 | — | — | — | 0.555 | |
| Hao et al.2018.12 | 84.1 | 96.6 | 99.1 | 0.127 | 0.053 | — | — | — | 0.555 | |
| Qi et al.2018.03 | 83.4 | 96 | 99 | 0.128 | 0.057 | — | — | — | 0.569 | |
| LiBackbone=ResNet-152, citation=[19]2019.01 | 83.2 | 96.5 | 98.9 | — | 0.187 | 0.54 | 0.134 | 0.095 | — | |
| MoukariBackbone=ResNet-2002019.01 | 83 | 96.6 | 99.3 | — | — | 0.569 | 0.133 | — | — | |
| Fu et al.Training samples=120K2018.03 | 82.8 | 96.5 | 99.2 | 0.115 | 0.051 | — | — | — | 0.509 | |
| Fu et al.2018.12 | 82.8 | 96.5 | 99.2 | 0.115 | 0.051 | — | — | — | 0.509 | |
| DORN2019.04 | 82.8 | 96.5 | 99.2 | 0.115 | 0.051 | — | — | — | 0.509 | |
| DORNSupervision=Supervised2019.10 | 82.8 | 96.5 | 99.2 | 0.115 | 0.051 | — | — | — | 0.509 | |
| ACANBackbone=ResNet-1012019.01 | 82.6 | 96.4 | 99 | — | 0.174 | 0.496 | 0.138 | 0.101 | — | |
| Li et al.# of Training=96K2018.05 | 82 | 96 | 98.9 | 0.139 | 0.058 | 0.505 | — | — | — | |
| LiBackbone=ResNet-101, citation=[19]2019.01 | 82 | 96 | 98.9 | — | — | 0.545 | 0.139 | — | — | |
| CaoBackbone=ResNet-1522019.01 | 81.9 | 96.5 | 99.2 | — | — | 0.54 | 0.141 | — | — | |
| TGVsamples=5002019.08 | 81.9 | 93 | 96.8 | 0.123 | — | 0.635 | — | — | — | |
| PAD-Net-ResNet50# of Training=7952018.05 | 81.7 | 95.4 | 98.7 | 0.12 | 0.055 | 0.582 | — | — | — | |
| Xu et al.Joint task learning=true2018.03 | 81.7 | 95.4 | 98.7 | 0.12 | 0.055 | — | — | — | 0.582 | |
| OursPost-processing=Gaussian smoothing2017.05 | 81.6 | 95 | 98.9 | — | — | — | — | — | — | |
| Lee et al.2018.03 | 81.5 | 96.3 | 99.1 | 0.139 | — | — | — | — | 0.572 | |
| ACANBackbone=ResNet-502019.01 | 81.5 | 96 | 98.9 | — | 0.18 | 0.518 | 0.144 | 0.11 | — | |
| YanBackbone=ResNet-1012019.01 | 81.3 | 96.5 | 99.3 | — | — | 0.502 | 0.135 | — | — | |
| LainaBackbone=ResNet-502017.05 | 81.1 | 95.3 | 98.8 | — | — | — | — | — | — | |
| Xu et al.# of Training=95K2018.05 | 81.1 | 95 | 98.6 | 0.121 | 0.052 | 0.586 | — | — | — | |
| Laina et al.2018.03 | 81.1 | 95.3 | 98.8 | 0.127 | 0.055 | — | — | — | 0.573 | |
| Xu et al.2018.03 | 81.1 | 95.4 | 98.7 | 0.121 | 0.052 | — | — | — | 0.586 | |
| LainaBackbone=ResNet-502019.01 | 81.1 | 95.3 | 98.8 | — | 0.195 | 0.573 | 0.127 | — | — | |
| XuBackbone=ResNet-502019.01 | 81.1 | 95.4 | 98.7 | — | — | 0.583 | 0.121 | — | — | |
| Laina et al.2018.12 | 81.1 | 95.3 | 98.8 | 0.127 | 0.055 | — | — | — | 0.573 | |
| MS-CRF2018.12 | 81.1 | 95.4 | 98.7 | 0.121 | 0.052 | — | — | — | 0.586 | |
| Laina et al.2019.04 | 81.1 | 95.3 | 98.8 | 0.127 | 0.055 | — | — | — | 0.573 | |
| MS-CRF2019.04 | 81.1 | 95.4 | 98.7 | 0.121 | 0.052 | — | — | — | 0.586 | |
| MS-CRFSupervision=Supervised2019.10 | 81.1 | 95.4 | 98.7 | 0.121 | 0.052 | — | — | — | 0.586 | |
| Ma and Karaman2018.03 | 81 | 95.9 | 98.9 | 0.143 | — | — | — | — | — | |
| OursPost-processing=None2017.05 | 80.9 | 94.5 | 98.6 | — | — | — | — | — | — | |
| LiBackbone=ResNet-50, citation=[19]2019.01 | 80.8 | 95.7 | 98.5 | — | — | 0.601 | 0.147 | — | — | |
| Chakrabarti et al.2018.03 | 80.6 | 95.8 | 98.7 | 0.149 | — | — | — | — | 0.62 | |
| Laina et al.# of Training=96K2018.05 | 80.1 | 95 | 98.6 | 0.129 | 0.056 | 0.583 | — | — | — | |
| Xu et al.# of Training=4.7K2018.05 | 79.3 | 94.8 | 98.4 | 0.139 | 0.063 | 0.609 | — | — | — | |
| Li et al.2018.03 | 78.8 | 95.8 | 99.1 | 0.143 | 0.063 | — | — | — | 0.635 | |
| Li et al. [29]Supervision=Supervised2019.10 | 78.8 | 95.8 | 99.1 | 0.143 | 0.063 | — | — | — | 0.635 | |
| f_T (all-real)Supervision=Real image-depth pairs, Training data=all-real2018.08 | 77.9 | 94.3 | 98.3 | — | 0.199 | 0.556 | 0.157 | 0.125 | — | |
| Dharmasiri et al.Joint task learning=true2018.03 | 77.6 | 95.3 | 98.9 | 0.156 | — | — | — | — | 0.624 | |
| Eigen [10]2017.05 | 76.9 | 95 | 98.8 | — | — | — | — | — | — | |
| Eigen and Fergus# of Training=7952018.05 | 76.9 | 95 | 98.8 | 0.158 | — | 0.641 | — | — | — | |
| Eigen et al. (VGG)Supervision=Real image-depth pairs, Backbone=VGG2018.08 | 76.9 | 95 | 98.8 | — | 0.214 | 0.641 | 0.158 | 0.121 | — | |
| Eigen and FergusJoint task learning=true2018.03 | 76.9 | 95 | 98.8 | 0.158 | — | — | — | — | 0.641 | |
| Eigencitation=[5]2019.01 | 76.9 | 95 | 98.8 | — | 0.214 | 0.641 | 0.158 | 0.121 | — | |
| Eigen et al.2018.12 | 76.9 | 95 | 98.8 | 0.158 | — | — | — | — | 0.641 | |
| Jafari et al.# of Training=7952018.05 | 76.2 | 94.8 | 98.8 | 0.157 | 0.068 | 0.673 | — | — | — | |
| SF-NetSupervision=Unsupervised2019.10 | 67.4 | 90 | 96.8 | 0.208 | 0.086 | — | — | — | 0.712 | |
| Liu et al.Supervision=Real image-depth pairs2018.08 | 65 | 90.6 | 97.6 | — | — | 0.759 | 0.213 | — | — | |
| Liucitation=[23]2019.01 | 65 | 90.6 | 97.6 | — | — | 0.759 | 0.213 | — | — | |
| Liu et al. [30]Supervision=Supervised2019.10 | 65 | 90.6 | 97.6 | 0.213 | 0.087 | — | — | — | 0.759 | |
| Cao et al.2018.03 | 64.6 | 89.2 | 96.8 | 0.232 | 0.091 | — | — | — | 0.819 | |
| Li et al.2018.03 | 62.1 | 88.6 | 96.8 | 0.232 | 0.094 | — | — | — | 0.821 | |
| Li2019.01 | 62.1 | 88.6 | 96.8 | — | — | 0.821 | 0.232 | — | — | |
| Li et al. [26]Supervision=Supervised2019.10 | 62.1 | 88.6 | 96.8 | 0.232 | 0.094 | — | — | — | 0.821 | |
| Liu2017.05 | 61.4 | 88.3 | 97.1 | — | — | — | — | — | — | |
| Eigen [11]2017.05 | 61.4 | 88.8 | 97.2 | — | — | — | — | — | — | |
| Liu et al.# of Training=795, citation=[32]2018.05 | 61.4 | 88.3 | 97.5 | 0.23 | 0.095 | 0.824 | — | — | — | |
| Liu et al.2018.03 | 61.4 | 88.3 | 97.1 | 0.23 | 0.095 | — | — | — | 0.824 | |
| Liu2019.01 | 61.4 | 88.3 | 97.1 | — | — | 0.824 | 0.23 | — | — | |
| Eigen et al.# of Training=120K2018.05 | 61.1 | 88.7 | 97.1 | 0.215 | — | 0.907 | — | — | — | |
| Eigen et al. (Fine)Supervision=Real image-depth pairs, Protocol=Fine2018.08 | 61.1 | 88.7 | 97.1 | — | 0.285 | 0.907 | 0.215 | 0.212 | — | |
| Eigen et al.2018.03 | 61.1 | 88.7 | 97.1 | 0.215 | — | — | — | — | 0.907 | |
| Eigencitation=[6]2019.01 | 61.1 | 88.7 | 97.1 | — | 0.285 | 0.907 | 0.158 | 0.121 | — | |
| Joint HCRF# of Training=7952018.05 | 60.5 | 89 | 97 | 0.22 | 0.094 | 0.745 | — | — | — | |
| Wang et al.Supervision=Supervised2019.10 | 60.5 | 89 | 97 | 0.22 | — | — | — | — | 0.824 | |
| Ladicky2017.05 | 54.2 | 82.9 | 94 | — | — | — | — | — | — | |
| Ladicky et al.# of Training=7952018.05 | 54.2 | 82.9 | 94.1 | — | — | — | — | — | — | |
| Ladicky et al.Supervision=Real image-depth pairs2018.08 | 54.2 | 82.9 | 94 | — | — | — | — | — | — |