Baseline TensorRT
- depth-anything/Depth-Anything-V2-Small-hf
- FP16
- AGX Orin
- Labels
- Image
- Depth
- Build
- Baseline TensorRT
- Precision
- FP16
- Runtime
- TensorRT 10.16.2.10
- Input size
- 518×518
- Output
- Relative inverse depth
Depth Anything V2 Small is a monocular depth estimator: from a single photo it predicts how near or far every pixel is, as relative depth.
Methodology
- Latency: median GPU time per image, batch 1, timed with CUDA events around each run.
- Accuracy: AbsRel and δ1 on a fixed Hypersim test subset of 32 images from 8 scenes, each depth map scale-and-shift aligned to the ground truth.
- Memory: peak memory while running the model.