Depth Anything 3
ByteDance's Depth Anything 3: depth from one image, with relative, metric and multi-view versions. The licence differs by variant, see the table.
stable Fork of ByteDance-Seed/Depth-Anything-3@3d835ec · backends: portable, kernels
- Paper Depth Anything 3: Recovering the Visual Space from Any Views arXiv · first released 13 Nov 2025
Paper 2025 Added in 0.1.0 Evaluated with 0.2.0
Use it
pip install fastforks Lean install with only what this model needs: pip install fastforks-core
from fastforks import depth
model = depth.load("da3mono-large") # kernels on a supported GPU, else portable PyTorch
result = model(image)
result.depth # also: confidence, intrinsics, mask, sky
result.backend # what ran: mode, device, dtype, kernels fastforks bench da3mono-large Times the model on this machine. load(…, mode="portable") or mode="kernels" picks a backend; usage guide.
Speed and memory
Every backend on one GPU (RTX 5060 Ti 16 GB) and the same input, the authors’ code included. Kernels runs fp16. Portable runs fp16 on CUDA.
| Backend | dtype | Time per call (ms) | vs original fp16 | Peak memory allocated / reserved | Load (s) | First call (s) |
|---|---|---|---|---|---|---|
| Original fp32float32 (the authors' autocast off) | fp32 | 127 | 0.67× | 2,419 MiB/2,450 MiB | 1.8 | 0.13 |
| Original fp16original default (api.py): torch.autocast(bfloat16) | fp16 | 85.2 | – | 2,997 MiB/3,052 MiB | 3.0 | 0.28 |
| Portable fp32 | fp32 | not measured | – | – | – | – |
| Portable | fp16 | 24.1 | 3.5× | 814 MiB/868 MiB | 2.4 | 0.03 |
| Kernels | fp16 | 14.9 | 5.7× | 805 MiB/868 MiB | 2.2 | 0.02 |
| Portable + torch.compile | 21.2 | 4.0× | 792 MiB/828 MiB | 25 (build) | – | |
| TensorRT fp16 | fp16 | 14.4 | 5.9× | 754 MiBenginenot comparable | 52 (build) | – |
Measured on RTX 5060 Ti 16 GB, torch 2.14.1+cu130, 2026-11-03 to 2026-11-16. First call, 3 warm-up, peak reset, median of 3 rounds x median of 5. Ratios divide two
times of this table (same GPU, same input). Original fp16 is the authors’ own precision option, else torch.autocast. torch.compile and TensorRT: build time instead of load time. TensorRT memory (engine, not comparable) is the GPU memory the engine took, read from the driver because TensorRT allocates outside PyTorch, so it can't be compared with the PyTorch peaks of the other rows.
Other devices
| Device | Backend | dtype | ms / call | Peak memory | Output vs reference | Input |
|---|---|---|---|---|---|---|
| Apple M1 Pro 16 GB MPS, torch 2.14.1, 2026-11-13 | Original | fp32 | 271 | 3,393 MiB process | matches | 994x1482 |
| Portable | fp16 | 280 | 2,499 MiB process | matches | 994x1482 | |
| Portable | fp32 | 293 | 2,517 MiB process | matches | 994x1482 | |
| Kernels | fp16 | not implemented | – | 994x1482 | ||
| NVIDIA L4 CUDA, torch 2.14.1+cu130, 2026-11-07 | Portable | fp16 | 22.6 | 787 MiB/828 MiB | – | 994x1482 |
| Kernels | fp16 | 19.2 | 783 MiB/828 MiB | – | 994x1482 | |
| NVIDIA L40S CUDA, torch 2.14.1+cu130, 2026-11-07 | Portable | fp16 | 7.80 | 787 MiB/828 MiB | – | 994x1482 |
| Kernels | fp16 | 7.66 | 783 MiB/828 MiB | – | 994x1482 | |
| NVIDIA A10 CUDA, torch 2.14.1+cu130, 2026-11-07 | Portable | fp16 | 22.0 | 787 MiB/828 MiB | – | 994x1482 |
| Kernels | fp16 | 18.5 | 783 MiB/828 MiB | – | 994x1482 | |
| NVIDIA A100-SXM4-40GB CUDA, torch 2.14.1+cu130, 2026-11-08 | Portable | fp16 | 15.5 | 787 MiB/828 MiB | – | 994x1482 |
| Kernels | fp16 | 12.8 | 783 MiB/828 MiB | – | 994x1482 | |
| NVIDIA RTX PRO 6000 CUDA, torch 2.14.1+cu130, 2026-11-08 | Portable | fp16 | 5.94 | 811 MiB/868 MiB | – | 994x1482 |
| Kernels | fp16 | 5.81 | 807 MiB/848 MiB | – | 994x1482 | |
CUDA memory is torch’s peak allocated / reserved; Apple has no such counter, so the Mac rows show the process footprint (weights, Python and torch included). “Output vs reference” compares the device’s output with the stored golden output under the card’s tolerance; – = not checked on that device. Kernels “not implemented” = no kernels for this model on that backend yet: it runs portable there.
Licence and weights
Commercial use allowed
- Code
- Apache-2.0 LICENSE
- Weights
- Apache-2.0 source
- Training data
- Public academic datasets only (model cards); teacher-student training on synthetic and real multi-view data per the paper
- The Mono and Metric weights are Apache-2.0 but their training data terms are unchecked (a note only); the other variants' weights are CC-BY-NC-4.0 (`no`).
Our reading of the licences, not legal advice.
Weights
DA3MONO-LARGE/model.safetensors 1.3 GB· hosted on Hugging Facefastforks downloads and checks these files on first use (fastforks download da3mono-large).
Accuracy
The test set through the authors’ code and through ours: same checkpoint, same protocol. The distance between the original’s fp32 and fp16 runs is what half precision alone costs; our runs should stay within it.
NYUv2 test
AbsRel / delta1
| Run | dtype | AbsRel ↓ | δ1 ↑ | Δ AbsRel vs fp32 original | images | ms / image |
|---|---|---|---|---|---|---|
| Original fp32 | fp32 | 0.0402 | 0.9821 | reference | 654 | 94.2 |
| Original fp16 | bf16 | 0.0402 | 0.9820 | ±0.0000 | 654 | 43.0 |
| Portable fp32 | fp32 | 0.0402 | 0.9821 | ±0.0000 | 654 | 92.0 |
| Portable | fp16 | 0.0402 | 0.9821 | ±0.0000 | 654 | 26.9 |
| Kernels | fp16 | 0.0402 | 0.9821 | ±0.0000 | 654 | 16.2 |
AbsRel / delta1 after a least-squares scale + shift fit (in disparity space for inverse-depth models), mean over images. Original = the authors’ code, same checkpoint and protocol. Measured on original fp32, original fp16: RTX 5060 Ti, torch 2.14.1+cu130, 2026-11-06; portable fp32: RTX 5060 Ti, torch 2.14.1+cu130, fastforks 6d3bab64, 2026-11-17; portable, kernels: RTX 5060 Ti, torch 2.14.1+cu130, fastforks 468fba10, 2026-11-06.
iBims-1
AbsRel / delta1
| Run | dtype | AbsRel ↓ | δ1 ↑ | Δ AbsRel vs fp32 original | images | ms / image |
|---|---|---|---|---|---|---|
| Original fp32 | fp32 | 0.0311 | 0.9872 | reference | 100 | 93.4 |
| Original fp16 | bf16 | 0.0311 | 0.9872 | ±0.0000 | 100 | 42.5 |
| Portable fp32 | fp32 | 0.0311 | 0.9872 | ±0.0000 | 100 | 91.6 |
| Portable | fp16 | 0.0311 | 0.9872 | ±0.0000 | 100 | 26.7 |
| Kernels | fp16 | 0.0311 | 0.9873 | ±0.0000 | 100 | 15.4 |
AbsRel / delta1 after a least-squares scale + shift fit (in disparity space for inverse-depth models), mean over images. Original = the authors’ code, same checkpoint and protocol. Measured on original fp32, original fp16: RTX 5060 Ti, torch 2.14.1+cu130, 2026-11-06; portable fp32: RTX 5060 Ti, torch 2.14.1+cu130, fastforks 6d3bab64, 2026-11-17; portable, kernels: RTX 5060 Ti, torch 2.14.1+cu130, fastforks 468fba10, 2026-11-06.
ETH3D two-view train left
AbsRel / delta1training data not stated: possibly seen
| Run | dtype | AbsRel ↓ | δ1 ↑ | Δ AbsRel vs fp32 original | images | ms / image |
|---|---|---|---|---|---|---|
| Original fp32 | fp32 | 0.0631 | 0.9567 | reference | 27 | 71.6 |
| Original fp16 | bf16 | 0.0629 | 0.9583 | +0.0002 | 27 | 38.0 |
| Portable fp32 | not measured | |||||
| Portable | fp16 | 0.0630 | 0.9568 | +0.0001 | 27 | 22.2 |
| Kernels | fp16 | 0.0631 | 0.9567 | ±0.0000 | 27 | 12.7 |
AbsRel / delta1 after a least-squares scale + shift fit (in disparity space for inverse-depth models), mean over images. Original = the authors’ code, same checkpoint and protocol. Measured on original fp32, original fp16: RTX 5060 Ti, torch 2.14.1+cu130, 2026-11-06; portable, kernels: RTX 5060 Ti, torch 2.14.1+cu130, fastforks 468fba10, 2026-11-06.
Middlebury 2014 trainingH left
AbsRel / delta1training data not stated: possibly seen
| Run | dtype | AbsRel ↓ | δ1 ↑ | Δ AbsRel vs fp32 original | images | ms / image |
|---|---|---|---|---|---|---|
| Original fp32 | fp32 | 0.0254 | 0.9954 | reference | 15 | 129 |
| Original fp16 | bf16 | 0.0253 | 0.9954 | +0.0001 | 15 | 82.6 |
| Portable fp32 | not measured | |||||
| Portable | fp16 | 0.0254 | 0.9954 | ±0.0000 | 15 | 27.8 |
| Kernels | fp16 | 0.0254 | 0.9954 | ±0.0000 | 15 | 15.5 |
AbsRel / delta1 after a least-squares scale + shift fit (in disparity space for inverse-depth models), mean over images. Original = the authors’ code, same checkpoint and protocol. Measured on original fp32, original fp16: RTX 5060 Ti, torch 2.14.1+cu130, 2026-11-06; portable, kernels: RTX 5060 Ti, torch 2.14.1+cu130, fastforks 468fba10, 2026-11-06.
Closeness to the original
Beyond the test-set scores: the raw output of ours against the original’s fp32 output on the same input, pixel by pixel. CPU, float32: max |diff| <= 0.005 vs the original. GPU, float16 (the default), portable and kernels: mean |diff| <= 0.5 % and p99 <= 2 % of the reference's median |value| (absolute error of the output depth).
| Check | Tolerance | Measured |
|---|---|---|
| CPU fp32, max |diff| | ≤ 0.005 (relative) | not recorded |
| GPU default dtypes, mean |diff| | ≤ 0.005 (relative) | 0.00032 (relative) |
| GPU default dtypes, p99 |diff| | ≤ 0.02 (relative) | 0.00078 (relative) |
Recorded for da3mono-large on RTX 5060 Ti, torch 2.14.1+cu130, 2026-11-12, with these kernels: dpt_fusedfastbb
Details
Depth Anything 3 (Lin et al. 2025, ByteDance): monocular depth from a DINOv2 ViT with DPT heads; relative (Mono), metric (Metric), any-view depth with camera (Large, Giant) and Giant aligned to Metric (Nested).
- Input / output
- Input: one RGB image. Output: a depth or inverse-depth map.
What the numbers mean, and which models need the camera focal length
kind(a card field,result.kind) says what the numbers mean:metric= metres;scale_invariant,affine_invariant,affine_invariant_log(log depth) andaffine_invariant_inverseare relative (unknown scale or shift, not metres). Metric models that use the focal length takemodel(image, intrinsics=K)(K[3, 3], pixels of the input):metric3d-vit-largeandda3metric-largeneed it for correct metres (da3metric-large: metres = output x focal / 300; withoutintrinsics=a 300 px focal length is assumed, wrong for a real camera).unidepth-v2-vitl14,unik3d-vitl,depth-proand MoGe estimate the camera themselves (givenintrinsics=replacefx);zoedepth-nkand the relative models (da3mono-largetoo) take none. Normals-only models (kindnone) returnresult.normals[B, H, W, 3](unit vectors, camera frame: x right, y down, z forward). - Input size
- any H x W >= 8, resized or padded inside the model (ViT patch 14); tested up to 3000x4000 (1,896 MiB allocated)
- Options
long_side=504>= 14· changes the output
- Outputs
depth,confidence,intrinsics,mask,sky(on request)- Checkpoint
arch="mono"- CUDA kernels
fastbbfast backbone kernels: flash attention, window attention, instance norm and rotary embeddings for ViT and similar backbonesdpt_fusedfused fp16 matrix multiplies and convolutions for DINOv2 backbones and DPT depth heads
- Apple (MPS)
- fp32, fp16
Full reference, generated from the repository
ByteDance's Depth Anything 3: depth from one image, with relative, metric and multi-view versions. The licence differs by variant, see the table.
Commercial use: da3mono-large, da3metric-large: yes; da3-large-1.1, da3-giant-1.1, da3nested-giant-large-1.1: no (weights: CC-BY-NC-4.0).
| variant | input | backend | time (ms) | peak memory (MiB, allocated / reserved) |
|---|---|---|---|---|
da3mono-large | synthetic (994x1482) | original (fp32) original portable kernels portable + torch.compile TensorRT fp16 | 127 85.2 24.1 14.9 21.2 14.4 | 2419 / 2450 2997 / 3052 814 / 868 805 / 868 792 / 828 1463 / 1520 |
da3metric-large | synthetic (994x1482) | original (fp32) original portable kernels portable + torch.compile TensorRT fp16 | 128 85.1 24.3 15.0 21.2 14.4 | 2418 / 2450 2996 / 3050 811 / 868 802 / 848 792 / 828 1465 / 1502 |
da3-large-1.1 | synthetic (994x1482) | original (fp32) original portable kernels portable + torch.compile TensorRT fp16 | 208 166 30.0 17.3 26.5 13.8 | 2797 / 2872 3374 / 3472 988 / 1088 891 / 950 841 / 868 1562 / 1622 |
da3-giant-1.1 | synthetic (994x1482) | original (fp32) original portable kernels portable + torch.compile TensorRT fp16 | 410 227 71.5 45.2 64.3 37.2 | 6428 / 6496 8590 / 8718 2634 / 2728 2535 / 2590 2487 / 2528 4795 / 4842 |
da3nested-giant-large-1.1 | synthetic (994x1482) | original (fp32) original portable kernels portable + torch.compile TensorRT fp16 | 533 308 94.9 59.7 85.4 51.9 | 7702 / 7760 10332 / 10402 3271 / 3374 3179 / 3238 3124 / 3194 6152 / 6226 |
Details
- Original code: https://github.com/ByteDance-Seed/Depth-Anything-3 at
3d835ec - Code licence: Apache-2.0 (permissive); files:
LICENSE - Weights licence: Apache-2.0
da3-large-1.1: CC-BY-NC-4.0da3-giant-1.1: CC-BY-NC-4.0da3nested-giant-large-1.1: CC-BY-NC-4.0
- Licence evidence: https://github.com/ByteDance-Seed/Depth-Anything-3/blob/main/README.md (Model Cards table) and the Hugging Face model cards (DA3-LARGE-1.1's card says Apache 2.0, the repository table CC BY-NC 4.0; the table is recorded)
- Training data: Public academic datasets only (model cards); teacher-student training on synthetic and real multi-view data per the paper
- Licence note: The Mono and Metric weights are Apache-2.0 but their training data terms are unchecked (a note only); the other variants' weights are CC-BY-NC-4.0 (
no). - Options:
long_side= 504 (>= 14; alsoprocess_res) - Kernel libraries:
fastbb,dpt_fused - Variants:
da3mono-large,da3metric-large,da3-large-1.1,da3-giant-1.1,da3nested-giant-large-1.1 - Outputs: depth, confidence, intrinsics, mask, sky (in
extras); kind:metric - Input size: any H x W >= 8, resized or padded inside the model (ViT patch 14)
- Tested up to (portable, fp16, RTX 5060 Ti 16 GB):
da3mono-large3000x4000 (1.9 GiB peak allocated, 2.1 reserved);da3metric-large3000x4000 (1.9 GiB peak allocated, 2.1 reserved);da3-large-1.13000x4000 (1.9 GiB peak allocated, 2.2 reserved);da3-giant-1.13000x4000 (5.1 GiB peak allocated, 5.4 reserved);da3nested-giant-large-1.13000x4000 (6.3 GiB peak allocated, 6.7 reserved) da3mono-large994x1482, first call: portable + torch.compile 25 s (cold Inductor cache), TensorRT fp16 52 s (ONNX + engine).da3metric-large994x1482, first call: portable + torch.compile 27 s (cold Inductor cache), TensorRT fp16 58 s (ONNX + engine).da3-large-1.1994x1482, first call: portable + torch.compile 45 s (cold Inductor cache), TensorRT fp16 63 s (ONNX + engine).da3-giant-1.1994x1482, first call: portable + torch.compile 47 s (cold Inductor cache), TensorRT fp16 84 s (ONNX + engine).da3nested-giant-large-1.1994x1482, first call: portable + torch.compile 60 s (cold Inductor cache), TensorRT fp16 154 s (ONNX + engine).da3mono-largeoriginal (fp32) 994x1482: float32 (the authors' autocast off)da3mono-largeoriginal 994x1482: original default (api.py): torch.autocast(bfloat16)da3metric-largeoriginal (fp32) 994x1482: float32 (the authors' autocast off)da3metric-largeoriginal 994x1482: original default (api.py): torch.autocast(bfloat16)da3-large-1.1original (fp32) 994x1482: float32 (the authors' autocast off)da3-large-1.1original 994x1482: original default (api.py): torch.autocast(bfloat16)da3mono-largeload / first call: original 2.96 s / 0.282 s, portable 2.4 s / 0.209 s (504x504), kernels 2.23 s / 0.17 s (504x504)da3metric-largeload / first call: original 2.7 s / 0.283 s, portable 4.84 s / 0.243 s (518x770), kernels 2.34 s / 0.215 s (518x770)da3-large-1.1load / first call: original 3.46 s / 0.45 s, portable 1.59 s / 0.408 s (518x770), kernels 2.16 s / 0.26 s (518x770)da3-giant-1.1load / first call: original 12.39 s / 0.684 s, portable 7.21 s / 0.357 s (518x770), kernels 5.38 s / 0.369 s (518x770)da3nested-giant-large-1.1load / first call: original 28.67 s / 1.626 s, portable 18.24 s / 0.455 s (518x770), kernels 7.35 s / 0.184 s (518x770)- Protocol, median of 5 after 3 warm-up, uint8 CUDA input (synthetic for bench_inputs sizes), card defaults; gc before each load; peak = max over the first call and the timed calls
Measured
AbsRel / delta1 after a least-squares scale + shift fit (in disparity space for inverse-depth models), mean over images. Original = the original code named under Details, same checkpoint and protocol.
| variant | dataset | original fp32 | original fp16 | portable | portable fp32 | kernels | note |
|---|---|---|---|---|---|---|---|
da3mono-large | NYUv2 test, AbsRel / delta1 | 0.0402 / 0.9821 | 0.0402 / 0.9820 | 0.0402 / 0.9821 | 0.0402 / 0.9821 | 0.0402 / 0.9821 | |
da3mono-large | iBims-1, AbsRel / delta1 | 0.0311 / 0.9872 | 0.0311 / 0.9872 | 0.0311 / 0.9872 | 0.0311 / 0.9872 | 0.0311 / 0.9873 | |
da3mono-large | ETH3D two-view train left, AbsRel / delta1 | 0.0631 / 0.9567 | 0.0629 / 0.9583 | 0.0630 / 0.9568 | 0.0631 / 0.9567 | possibly in the training data | |
da3mono-large | Middlebury 2014 trainingH left, AbsRel / delta1 | 0.0254 / 0.9954 | 0.0253 / 0.9954 | 0.0254 / 0.9954 | 0.0254 / 0.9954 | possibly in the training data | |
da3metric-large | NYUv2 test, AbsRel / delta1 | 0.0398 / 0.9820 | 0.0398 / 0.9820 | 0.0398 / 0.9820 | 0.0398 / 0.9820 | 0.0398 / 0.9820 | |
da3metric-large | iBims-1, AbsRel / delta1 | 0.0328 / 0.9864 | 0.0328 / 0.9863 | 0.0328 / 0.9864 | 0.0328 / 0.9864 | 0.0328 / 0.9864 | |
da3metric-large | ETH3D two-view train left, AbsRel / delta1 | 0.0716 / 0.9343 | 0.0716 / 0.9343 | 0.0716 / 0.9342 | 0.0717 / 0.9342 | possibly in the training data | |
da3metric-large | Middlebury 2014 trainingH left, AbsRel / delta1 | 0.0274 / 0.9949 | 0.0275 / 0.9949 | 0.0274 / 0.9949 | 0.0274 / 0.9949 | possibly in the training data | |
da3-large-1.1 | NYUv2 test, AbsRel / delta1 | 0.0468 / 0.9752 | 0.0468 / 0.9752 | 0.0468 / 0.9752 | 0.0468 / 0.9752 | 0.0468 / 0.9752 | |
da3-large-1.1 | iBims-1, AbsRel / delta1 | 0.0413 / 0.9763 | 0.0413 / 0.9764 | 0.0413 / 0.9763 | 0.0413 / 0.9763 | 0.0413 / 0.9764 | |
da3-large-1.1 | ETH3D two-view train left, AbsRel / delta1 | 0.0867 / 0.9177 | 0.0869 / 0.9176 | 0.0868 / 0.9177 | 0.0867 / 0.9177 | 0.0867 / 0.9177 | possibly in the training data |
da3-large-1.1 | Middlebury 2014 trainingH left, AbsRel / delta1 | 0.0418 / 0.9901 | 0.0419 / 0.9899 | 0.0418 / 0.9901 | 0.0418 / 0.9901 | 0.0419 / 0.9900 | possibly in the training data |
da3-giant-1.1 | NYUv2 test, AbsRel / delta1 | 0.0397 / 0.9794 | 0.0397 / 0.9794 | 0.0397 / 0.9794 | 0.0397 / 0.9794 | 0.0397 / 0.9794 | |
da3-giant-1.1 | iBims-1, AbsRel / delta1 | 0.0318 / 0.9819 | 0.0317 / 0.9820 | 0.0318 / 0.9820 | 0.0318 / 0.9819 | 0.0318 / 0.9819 | |
da3-giant-1.1 | ETH3D two-view train left, AbsRel / delta1 | 0.0705 / 0.9416 | 0.0704 / 0.9415 | 0.0704 / 0.9415 | 0.0705 / 0.9416 | 0.0704 / 0.9420 | possibly in the training data |
da3-giant-1.1 | Middlebury 2014 trainingH left, AbsRel / delta1 | 0.0284 / 0.9930 | 0.0284 / 0.9930 | 0.0284 / 0.9930 | 0.0284 / 0.9930 | 0.0284 / 0.9930 | possibly in the training data |
da3nested-giant-large-1.1 | NYUv2 test, AbsRel / delta1 | 0.0387 / 0.9809 | 0.0387 / 0.9810 | 0.0387 / 0.9809 | 0.0387 / 0.9809 | 0.0387 / 0.9809 | |
da3nested-giant-large-1.1 | iBims-1, AbsRel / delta1 | 0.0320 / 0.9825 | 0.0320 / 0.9826 | 0.0320 / 0.9824 | 0.0320 / 0.9825 | 0.0320 / 0.9825 | |
da3nested-giant-large-1.1 | ETH3D two-view train left, AbsRel / delta1 | 0.0761 / 0.9361 | 0.0762 / 0.9346 | 0.0761 / 0.9362 | 0.0760 / 0.9362 | possibly in the training data | |
da3nested-giant-large-1.1 | Middlebury 2014 trainingH left, AbsRel / delta1 | 0.0298 / 0.9925 | 0.0299 / 0.9925 | 0.0298 / 0.9925 | 0.0298 / 0.9925 | possibly in the training data |
Measured on portable: NVIDIA GeForce RTX 5060 Ti, torch 2.14.1+cu130, fastforks 468fba10, 2026-11-05 to 2026-11-06; portable32: NVIDIA GeForce RTX 5060 Ti, torch 2.14.1+cu130, fastforks 468fba10 / 6d3bab64, 2026-11-05 to 2026-11-17; kernels: NVIDIA GeForce RTX 5060 Ti, torch 2.14.1+cu130, fastforks 468fba10, 2026-11-05 to 2026-11-06.
Parity
CPU, float32: max |diff| <= 0.005 vs the original. GPU, float16 (the default), portable and kernels: mean |diff| <= 0.5 % and p99 <= 2 % of the reference's median |value| (absolute error of the output depth).
| variant | kernels error vs original, mean / p99 | GPU, torch | kernels used |
|---|---|---|---|
da3mono-large | 0.00032 / 0.00078 | RTX 5060 Ti, torch 2.14.1+cu130 | dpt_fused@48e0065bb7cf:sm_120, fastbb@7d058322634d:sm_120 |
da3metric-large | 0.00196 / 0.00259 | RTX 5060 Ti, torch 2.14.1+cu130 | dpt_fused@48e0065bb7cf:sm_120, fastbb@7d058322634d:sm_120 |
da3-large-1.1 | 0.00011 / 0.00031 | RTX 5060 Ti, torch 2.14.1+cu130 | dpt_fused@48e0065bb7cf:sm_120, fastbb@7d058322634d:sm_120 |
da3-giant-1.1 | 0.00014 / 0.00037 | RTX 5060 Ti, torch 2.14.1+cu130 | dpt_fused@48e0065bb7cf:sm_120, fastbb@7d058322634d:sm_120 |
da3nested-giant-large-1.1 | 0.00151 / 0.00184 | RTX 5060 Ti, torch 2.14.1+cu130 | dpt_fused@48e0065bb7cf:sm_120, fastbb@7d058322634d:sm_120 |
Example output
One fixed input through this checkpoint, the same for every model of the task (kernels · float16 · NVIDIA GeForce RTX 5060 Ti).
Input: A cup of coffee, Rachel Michetti (CC0).