The field of depth estimation is moving towards improving robustness and generalization under various environmental conditions. Researchers are developing new benchmarks and evaluation methods to assess the robustness of depth estimation models, such as procedural generation of 3D scenes with controlled perturbations. Furthermore, innovative approaches are being proposed to enhance the performance of monocular depth estimation models in challenging conditions, including adverse weather, illumination variations, and sensor-induced distortions. These approaches include unsupervised consistency regularization, spatial distance constraints, and robust feature encoders. Noteworthy papers in this area include:

• Evaluating Robustness of Monocular Depth Estimation with Procedural Scene Perturbations, which introduces a new benchmark for systematic robustness evaluation.
• Depth Anything at Any Condition, which presents a foundation model capable of handling diverse environmental conditions.
• RobuSTereo, which proposes a novel framework for robust zero-shot stereo matching under adverse weather conditions.