AdaBins: Depth Estimation Using Adaptive Bins

By Shariq Farooq Bhat, Ibraheem Alhashim, Peter Wonka. Computer Vision and Pattern Recognition, 2020

1,200 citations recorded by the source corpus.

We address the problem of estimating a high quality dense depth map from a single RGB input image. We start out with a baseline encoder-decoder convolutional neural network architecture and pose the question of how the global processing of information can help improve overall depth estimation. To this end, we propose a transformer-based architecture block that divides the depth range into bins whose center value is estimated adaptively per image. The final depth values are estimated as linear combinations of the bin centers. We call our new building block AdaBins. Our results show a decisive improvement over the state-of-the-art on several popular depth datasets across all metrics. We also validate the effectiveness of the proposed block with an ablation study and provide the code and corresponding pre-trained weights of the new state-of-the-art model.

Subfields: Computer Vision and Pattern Recognition, Media Technology

View Paper: https://www.semanticscholar.org/paper/9717f7d873859ca0080caf944f7a2d9b456be121

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

In Media Technology

Citations in the corpus

Depth Pro: Sharp Monocular Metric Depth in Less Than a Second (2024, neutral): “A line of work sought to improve metric depth prediction through a global distribution of depth values (Fu et al., 2018; Bhat et al., 2021; 2022; Li et al., 2024b) and further conditioning on scene type (Bhat et al., 2023).”
Learning to Upsample by Learning to Sample (2023, background): “Dense prediction refers to a branch of tasks that require point-wise label prediction, such as semantic/instance/panoptic segmentation [2, 39, 40, 8, 7, 13, 11, 16, 19], object detection [33, 4, 24, 36], and monocular depth estimation [38, 18, 3, 21].”
Repurposing Diffusion-Based Image Generators for Monocular Depth Estimation (2023, positive): “Some authors treat depth estimation as a combined regression-classification task, using various binning strategies like Ad-aBins [4] or BinsFormer [26] to discretize depth range.”
Unleashing Text-to-Image Diffusion Models for Visual Perception (2023, result): “Instead, we compare the available result with more common segmentation heads like Semantic FPN [24] and UperNet [54].”
VoxFormer: Sparse Voxel Transformer for Camera-Based 3D Semantic Scene Completion (2023, methodology): “We leverage off-the-shelf depth estimation models such as monocular depth [69] or stereo depth [70] to directly predict the depth Z(u, v) of each image pixel (u, v).”
BinsFormer: Revisiting Adaptive Bins for Monocular Depth Estimation (2022, methodology): “While Adabins [15] boosts depth estimation performance to a remarkable extent, several dilemmas still exist.”
Depth Anything: Unleashing the Power of Large-Scale Unlabeled Data (2024, background): “Following this, many works consistently improve the depth estimation accuracy by carefully designing the regression task as a classification task [3, 34], introducing more priors [32, 53, 75, 82], and better objective functions [67, 77], etc .”
Metric3D: Towards Zero-shot Metric 3D Prediction from A Single Image (2023, methodology): “Though monocular metric depth estimation methods [71, 4] work on a single dataset with a single camera model, they cannot generalize to unseen cameras or scenes.”
ZoeDepth: Zero-shot Transfer by Combining Relative and Metric Depth (2023, methodology): “[48] ResNeXt-101 114M BTS [20] DenseNet-161 47M AdaBins [5] EfﬁcientNet-B5 78M LocalBins [6] EfﬁcientNet-B5 74M NeWCRFs [50] Swin-L 270M ZoeDepth (S-L) Beit-L 345M Table 17.”
Delving into Deep Imbalanced Regression (2021, neutral): “Each pair is annotated by multiple annotators with an averaged Following standard practices (Bhat et al., 2020; Hu et al., 2019), we use 50K images for training and 654 images for testing.”
Metric3D v2: A Versatile Monocular Geometric Foundation Model for Zero-Shot Metric Depth and Surface Normal Estimation (2024, neutral): “Personal use is permitted, but republication/redistribution requires IEEE permission.”
Depth Anything V2 (2024, neutral): “Early works [18, 19, 5] focus on the in-domain metric depth estimation, where training and test images must share the same domain [70, 24].”
DINOv2: Learning Robust Visual Features without Supervision (2023, neutral): “(2021). lin.”
MonoViT: Self-Supervised Monocular Depth Estimation with a Vision Transformer (2022, background): “Deep learning paradigms favoured the blooming of this latter approach [12, 24, 13, 3], at the cost of requiring extensive collections of images annotated with depth labels in order to carry out the training effectively.”
UniDepth: Universal Monocular Metric Depth Estimation (2024, methodology): “While existing MMDE methods [2, 13, 15, 33, 34, 36, 50] have demonstrated remarkable accuracy across different benchmarks, they require training and testing on datasets with similar camera intrinsics and scene scales.”
BEVDepth: Acquisition of Reliable Depth for Multi-view 3D Object Detection (2022, methodology): “When evaluating all foreground regions, the Base Detector only achieves 3.03 AbsRel, which is greatly poor than existing depth estimation algorithms (Li et al. 2022a; Bhat, Alhashim, and Wonka 2021).”
MonoScene: Monocular 3D Semantic Scene Completion (2021, methodology): “We use the pretrained AdaBin [4] to infer a depth map (x̂depth) serving as input for AICNetrgb.”
SurroundOcc: Multi-Camera 3D Occupancy Prediction for Autonomous Driving (2023, methodology): “We compare with several baseline methods on nuScenes dataset, which can be roughly classified as four categories: Depth estimation: SurroundDepth [57], AdaBins [4], NeWCRFs [64].”