Monocular depth estimation (MDE) has emerged as a cost-effective alternative to hardware-dependent 3D sensing modalities, yet predicting precise metric depth remains a persistent challenge due to intrinsic scale ambiguity, edge bleeding, and limited zero-shot generalization in unseen environments. Recent multi-modal approaches incorporate 1D/3D Single-Photon Avalanche Diode (SPAD) data to resolve these ambiguities, but they mainly treat SPAD noise as a harmful artifact required for denoising or leave the noise ignored. In this paper, we present an essentially different insight: native SPAD noise is an informative prior that implicitly encodes deterministic environmental characteristics. Driven by this paradigm shift, we introduce a novel, lightweight Test-Time Gated Attention framework by Harnessing SPAD Noise, therefore dubbed HSN. Tailored for indoor scenarios, HSN leverages the synergistic combination of 1D SPAD histograms and Gate Adapters (GAs). Through three complementary optimization phases, it generates channel-wise gated attention vectors to dynamically adjust encoder feature maps at inference time, effectively resolving ambiguous scale and boundary challenges without the heavy computational overhead or dense 3D modalities that other TTA methods suffer from. Extensive experiments on challenging real-world indoor benchmarks show that HSN achieves state-of-the-art metric depth accuracy in in-domain settings, while exhibiting even greater potential under zero-shot generalization to unfamiliar environments. Furthermore, interpretability analysis reveals that the proposed gated attention mechanism adaptively shifts attention toward deeper semantic layers under intensifying SPAD noise. Code will be made publicly available.
Figures & tables
Figure 1: Overview of the proposed Test-Time Gated Attention framework (HSN). (a) illustrates how Gate Adapter (GA) modules are embedded between the selected encoder map and its corresponding decoder map, where the targeted encoder map is routed to the subsequent decoding layer. Conversely, encoder maps not routed to decoding layers bypass GA modulation entirely. (b) details the architectural design of the GA module and the TTA workflow. During inference, the pre-trained encoder-decoder network θ remains frozen to preserve its inherent structural priors, restricting optimization strictly to the GA parameters ϕ . These modules leverage the SPAD noise signal to dynamically generate a channel-wise gated attention vector g , thereby calibrating the selected encoder maps prior to decoding.
Method
δ1↑
δ2↑
δ3↑
Abs Rel ↓
RMSE ↓
log10↓
Baselines
Eigen et al . [ 8 ]
0.769
0.950
0.988
0.158
0.641
–
DORN [ 9 ]
0.828
0.965
0.992
0.115
0.509
0.051
Hu et al . [ 16 ]
0.866
0.975
0.993
0.115
0.530
0.050
BTS [ 21 ]
0.885
0.978
0.994
0.110
0.392
0.047
LocalBins [ 4 ]
0.907
0.987
0.998
0.099
0.357
0.042
Table 1: Quantitative in-domain comparison on NYU Depth V2. re indicates our re-evaluation using the official codes and same environments as our method for a fair TTA comparison. ”N” indicates that the model is fine-tuned on NYU Depth V2, and ”L” denotes the Large backbone version. Bold and underline indicate the best and second best results respectively.
Table 3
Figure 2: Comparison on the NYU Depth V2 (left) and RGB-SPAD (right) datasets. Only the last sample on the right is a poster; all other samples are captured scenes.
Modification
In-Domain
Unseen
δ1↑
Abs Rel ↓
RMSE↓
δ1↑
Abs Rel↓
RMSE↓
Baseline (Original)
0.901
0.104
0.366
0.424
0.347
1.103
w/o ω
0.903
0.102
0.364
0.688
0.282
0.842
only Lphy
0.906
0.101
0.361
0.687
0.284
0.841
only Lfid
0.903
0.103
0.366
0.687
0.283
0.841
w/o ZoDA
-
-
-
0.427
0.332
1.080
Table 4: Effect of HSN components on AdaBins inference. Adabins is trained on NYU-Depth-v2 (Train). In-Domain performance is evaluated on NYU-Depth-v2 (Test), while zero-shot generalization (Unseen) is evaluated on the RGB-SPAD dataset. Bold and underline indicate the best and second best results respectively.
Strategy
In-Domain data
Unseen data
δ1↑
Abs Rel ↓
RMSE ↓
δ1↑
Abs Rel ↓
RMSE ↓
Baseline (Original)
0.901
0.104
0.366
0.424
0.347
1.103
In-Domain Strategy
0.903
0.102
0.364
0.427
0.333
1.081
Unseen Strategy
0.811
0.129
0.436
0.689
0.282
0.842
Table 5: Cross-strategy evaluation of data domains using AdaBins. We evaluate the compatibility between domain-specific strategies and data distributions across multiple depth metrics. Bold values indicate optimal matching, and arrows ( ↑ for δ1 , ↓ for others) denote the preferred metric directions, where performance degradation occurs under strategy-data mismatch.
Figure 3: Metric curves vs. optimization steps for NYU-Depth-V2 (left) and RGB-SPAD (right). We apply our method on AdaBins to observe the performance across 1 to 5 optimization steps.
Self-supervised monocular depth estimation remains challenging due to the loss of fine-grained details and limited adaptability to varying scene structures. In this paper, we propose SAP-Depth, a novel architecture based on Scene-Adaptive Probabilistic. Unlike traditional methods, we introduce a Dynamic Bins mechanism powered by a Learnable Parameter Predictor, which adaptively estimates scene-specific depth distributions conditioned on the input image context. Furthermore, a Bin-wise Probability Distribution Decoder is designed to effectively fuse multi-scale representations to predict dense depth probability distributions, significantly enhancing boundary sharpness and structural recovery. Through the collaborative coupling of these proposed modules, our model achieves dynamic prediction of scene-specific depth distributions. Extensive experiments on the KITTI dataset show that SAP-Depth outperforms existing state-of-the-art methods in accuracy, particularly in its ability to better preserve complex edge details. Further, tests on the Make3D dataset validate the model's excellent generalization capability and robustness.
Monocular depth estimation has seen remarkable progress through foundation models achieving robust zero-shot generalization, yet their computational demands place them far beyond the reach of embedded and mobile platforms. Lightweight alternatives exist, but have been developed almost exclusively within single-domain, self-supervised paradigms, failing silently under domain shift. We present ZipDepth, a compact monocular depth network that bridges this gap by combining an efficient reparameterizable encoder-decoder with large-scale knowledge distillation from a foundation model over a large multi-domain training set. Comprising just 6.1M parameters, ZipDepth runs at real-time rates from server GPUs to power-constrained devices, achieving the best trade-off between zero-shot accuracy and deployment efficiency among lightweight models across five benchmarks, taking a significant step towards the accuracy of foundation models with 50x more parameters.
Monocular depth estimation (MDE) typically produces depth estimations that are defined up to an unknown scale or shift. When only sparse metric anchors are available, recovering accurate metric depth becomes challenging yet necessary for practical applications. We address this problem by formulating metric depth recovery as image-adaptive scale field modeling. Instead of directly correcting the depth, we reformulate the correction as a low-dimensional linear combination of image-adaptive basis maps. These maps are derived from semantic and geometric cues encoded in the MDE estimations and intermediate representations. The weights of basis maps are efficiently determined from sparse metric anchors via a least-squares problem. This formulation yields improved metric depth accuracy, strong robustness under extreme anchor sparsity, and an interpretable decomposition of spatial scale variations. Extensive experiments across multiple datasets and representative MDE models demonstrate the effectiveness and general applicability of our approach.