PointDiT
Google's PointDiT: a diffusion transformer that denoises a 3D point map in three steps from one image, conditioned on frozen DINOv3 features. Code and weights are Apache-2.0; the DINOv3 encoder it loads is under the DINOv3 License.
experimental Fork of google-research/pointdit@11f53a3 · backends: portable, kernels
- Paper PointDiT: Pixel-Space Diffusion for Monocular Geometry Estimation ICML 2026 · first released 2 Jul 2026
Paper 2026 Added in 0.2.0 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("pointdit-b-512") # kernels on a supported GPU, else portable PyTorch
result = model(image)
result.points # also: depth
result.backend # what ran: mode, device, dtype, kernels fastforks bench pointdit-b-512 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 fp32upstream wild-image resize (~1024 tokens), 3 Euler steps | fp32 | 105 | 0.46× | 977 MiB/1,006 MiB | 4.9 | 0.28 |
| Original (own default)the authors' mixed-precision option (bfloat16); upstream wild-image resize (~1024 tokens), 3 Euler steps; authors bf16 autocast | bf16 | 48.7 | – | 1,214 MiB/1,246 MiB | 2.6 | 0.11 |
| Portable fp32 | fp32 | not measured | – | – | – | – |
| Portable | fp16 | 31.1 | 1.6× | 446 MiB/504 MiB | 1.7 | 0.03 |
| Kernels | fp16 | 20.5 | 2.4× | 462 MiB/630 MiB | 1.7 | 0.02 |
| Portable + torch.compile | 23.7 | 2.1× | 465 MiB/524 MiB | 34 (build) | – | |
| TensorRT fp16 | fp16 | 15.5 | 3.1× | 394 MiBenginenot comparable | 87 (build) | – |
Measured on RTX 5060 Ti 16 GB, torch 2.14.1+cu130, 2026-11-15 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, 2026-11-15 | Original | fp32 | 468 | 3,167 MiB process | – | 994x1482 |
| Portable | fp16 | 337 | 1,912 MiB process | matches | 994x1482 | |
| Portable | fp32 | 367 | 1,812 MiB process | matches | 994x1482 | |
| Kernels | fp16 | not implemented | – | 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
- LicenseRef-DINOv3-License source
- Training data
- Mixture of synthetic and real point-map datasets ("mixdata") per the PointDiT paper; the DINOv3 encoder: LVD-1689M (not released)
- The PointDiT code and weights are Apache-2.0; the frozen DINOv3 ViT-B/16 and ViT-L/16 encoders are under the DINOv3 License (Meta: attribution and conditions apply; gated on Hugging Face); read it before commercial use.
Our reading of the licences, not legal advice.
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 | 0.0453 | 0.9773 | reference | 654 | 109 |
| Original fp16 | bf16 | 0.0454 | 0.9773 | −0.0001 | 654 | 48.6 |
| Portable fp32 | fp32 | 0.0453 | 0.9773 | ±0.0000 | 654 | 103 |
| Portable | fp16 | 0.0453 | 0.9773 | ±0.0000 | 654 | 32.0 |
| Kernels | fp16 | 0.0453 | 0.9773 | ±0.0000 | 654 | 21.6 |
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, 2026-11-15; portable fp32: RTX 5060 Ti, torch 2.14.1+cu130, fastforks 6a5100ec, 2026-11-15; portable, kernels: RTX 5060 Ti, torch 2.14.1+cu130, fastforks 2fb984b3, 2026-11-15.
iBims-1
AbsRel / delta1
| Run | dtype | AbsRel ↓ | δ1 ↑ | Δ AbsRel vs fp32 original | images | ms / image |
|---|---|---|---|---|---|---|
| Original fp32 | fp32 | 0.0374 | 0.9821 | reference | 100 | 109 |
| Original fp16 | bf16 | 0.0374 | 0.9820 | ±0.0000 | 100 | 48.7 |
| Portable fp32 | fp32 | 0.0374 | 0.9821 | ±0.0000 | 100 | 103 |
| Portable | fp16 | 0.0374 | 0.9821 | ±0.0000 | 100 | 32.0 |
| Kernels | fp16 | 0.0374 | 0.9821 | ±0.0000 | 100 | 21.6 |
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, 2026-11-15; portable fp32: RTX 5060 Ti, torch 2.14.1+cu130, fastforks 6a5100ec, 2026-11-15; portable, kernels: RTX 5060 Ti, torch 2.14.1+cu130, fastforks 2fb984b3, 2026-11-15.
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 | 0.0896 | 0.9085 | reference | 27 | 108 |
| Original fp16 | bf16 | 0.0896 | 0.9086 | ±0.0000 | 27 | 49.3 |
| Portable fp32 | not measured | |||||
| Portable | fp16 | 0.0896 | 0.9086 | ±0.0000 | 27 | 31.4 |
| Kernels | fp16 | 0.0896 | 0.9085 | ±0.0000 | 27 | 25.0 |
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, 2026-11-15; portable, kernels: RTX 5060 Ti, torch 2.14.1+cu130, fastforks 2fb984b3, 2026-11-15.
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 | 0.0313 | 0.9948 | reference | 15 | 107 |
| Original fp16 | bf16 | 0.0313 | 0.9948 | ±0.0000 | 15 | 50.5 |
| Portable fp32 | not measured | |||||
| Portable | fp16 | 0.0313 | 0.9948 | ±0.0000 | 15 | 31.3 |
| Kernels | fp16 | 0.0313 | 0.9948 | ±0.0000 | 15 | 25.2 |
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, 2026-11-15; portable, kernels: RTX 5060 Ti, torch 2.14.1+cu130, fastforks 2fb984b3, 2026-11-15.
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| <= 1.5e-06 vs the original. GPU, float16 (the default), portable and kernels: mean |diff| <= 0.00029 and p99 <= 0.0012 (absolute error of the output points).
| Check | Tolerance | Measured |
|---|---|---|
| CPU fp32, max |diff| | ≤ 0.0000015 | not recorded |
| GPU default dtypes, mean |diff| | ≤ 0.00029 | 0.00015 |
| GPU default dtypes, p99 |diff| | ≤ 0.0012 | 0.00056 |
Recorded for pointdit-b-512 on RTX 5060 Ti, torch 2.14.1+cu130, 2026-11-14, with these kernels: dpt_fusedfastbb
Details
PointDiT (Xu et al., Google 2026): monocular geometry as generation; a pixel-space diffusion transformer denoises a point map in 3 Euler steps, conditioned on frozen DINOv3 features of the image. Outputs an affine-invariant point map and its depth.
- 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 >= 1, resized or padded inside the model; tested up to 3000x4000 (843 MiB allocated)
- Options
tokens=1024>= 1· changes the outputsteps=3>= 1· changes the output
- Outputs
points,depth- Checkpoint
arch="b",encoder="dinov3-vitb16"- 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
Limitations
- Affine-invariant output: points up to an unknown scale and 3-D shift, depth (their z) up to scale and shift; align before use.
- Trained at 512 x 512: inputs are resized to about `tokens` 16-pixel patches at their aspect ratio (bilinear) and the point map is resized back with nearest, as the original's wild-image path.
- Kernels mode captures CUDA graphs: the graph's private memory pool and static outputs stay reserved after the call, so peak reserved is above portable by ~85 MiB (b), more for l (peak allocated is equal or lower); `cuda_graphs=false` or `release_graphs()` frees it at the cost of launch overhead.
Full reference, generated from the repository
Google's PointDiT: a diffusion transformer that denoises a 3D point map in three steps from one image, conditioned on frozen DINOv3 features. Code and weights are Apache-2.0; the DINOv3 encoder it loads is under the DINOv3 License.
Commercial use: yes.
| variant | input | backend | time (ms) | peak memory (MiB, allocated / reserved) |
|---|---|---|---|---|
pointdit-b-512 | Middlebury 2014 half (994x1482) | original (fp32) original portable kernels portable + torch.compile TensorRT fp16 | 105 48.7 31.1 20.5 23.7 15.5 | 977 / 1006 1214 / 1246 446 / 504 462 / 630 465 / 524 795 / 898 |
pointdit-l-512 | Middlebury 2014 half (994x1482) | original (fp32) original portable kernels portable + torch.compile TensorRT fp16 | 335 112 94.6 61.3 76.0 49.1 | 3092 / 3148 3953 / 4010 1297 / 1430 1310 / 1676 1316 / 1430 2474 / 2626 |
Details
- Original code: https://github.com/google-research/pointdit at
11f53a30aeb4 - Code licence: Apache-2.0 (permissive); files:
LICENSE - Weights licence: LicenseRef-DINOv3-License
- Licence evidence: https://github.com/google-research/pointdit/blob/11f53a30aeb4d34d49d2beda655ee6f19b2ad7f8/LICENSE (Apache-2.0); https://huggingface.co/haofeixu/pointdit (license: apache-2.0); encoder: https://huggingface.co/facebook/dinov3-vitl16-pretrain-lvd1689m (DINOv3 License, gated)
- Training data: Mixture of synthetic and real point-map datasets ("mixdata") per the PointDiT paper; the DINOv3 encoder: LVD-1689M (not released)
- Licence note: The PointDiT code and weights are Apache-2.0; the frozen DINOv3 ViT-B/16 and ViT-L/16 encoders are under the DINOv3 License (Meta: attribution and conditions apply; gated on Hugging Face); read it before commercial use.
- Options:
tokens= 1024 (>= 1),steps= 3 (>= 1) - Kernel libraries:
fastbb,dpt_fused - Variants:
pointdit-b-512,pointdit-l-512 - Outputs: points, depth; kind:
affine_invariant - Input size: any H x W >= 1, resized or padded inside the model
- Tested up to (portable, fp16, RTX 5060 Ti 16 GB):
pointdit-b-5123000x4000 (0.8 GiB peak allocated, 1.0 reserved);pointdit-l-5123000x4000 (1.6 GiB peak allocated, 2.2 reserved) pointdit-b-512994x1482, first call: portable + torch.compile 34 s (cold Inductor cache), TensorRT fp16 86 s (ONNX + engine).pointdit-l-512994x1482, first call: portable + torch.compile 57 s (cold Inductor cache), TensorRT fp16 150 s (ONNX + engine).pointdit-b-512original (fp32) 994x1482: upstream wild-image resize (~1024 tokens), 3 Euler stepspointdit-b-512original 994x1482: the authors' mixed-precision option (bfloat16); upstream wild-image resize (~1024 tokens), 3 Euler steps; authors bf16 autocastpointdit-l-512original (fp32) 994x1482: upstream wild-image resize (~1024 tokens), 3 Euler stepspointdit-l-512original 994x1482: the authors' mixed-precision option (bfloat16); upstream wild-image resize (~1024 tokens), 3 Euler steps; authors bf16 autocastpointdit-b-512load / first call: original 2.57 s / 0.111 s, portable 1.67 s / 0.184 s (518x770), kernels 1.74 s / 0.155 s (518x770)pointdit-l-512load / first call: original 9.47 s / 0.119 s, portable 3.96 s / 0.247 s (518x770), kernels 3.91 s / 0.234 s (518x770)- Protocol, first call, 3 warm-up, peak reset, median of 3 rounds x median of 5
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 |
|---|---|---|---|---|---|---|---|
pointdit-b-512 | NYUv2 test, AbsRel / delta1 | 0.0453 / 0.9773 | 0.0454 / 0.9773 | 0.0453 / 0.9773 | 0.0453 / 0.9773 | 0.0453 / 0.9773 | |
pointdit-b-512 | iBims-1, AbsRel / delta1 | 0.0374 / 0.9821 | 0.0374 / 0.9820 | 0.0374 / 0.9821 | 0.0374 / 0.9821 | 0.0374 / 0.9821 | |
pointdit-b-512 | ETH3D two-view train left, AbsRel / delta1 | 0.0896 / 0.9085 | 0.0896 / 0.9086 | 0.0896 / 0.9086 | 0.0896 / 0.9085 | possibly in the training data | |
pointdit-b-512 | Middlebury 2014 trainingH left, AbsRel / delta1 | 0.0313 / 0.9948 | 0.0313 / 0.9948 | 0.0313 / 0.9948 | 0.0313 / 0.9948 | possibly in the training data | |
pointdit-l-512 | NYUv2 test, AbsRel / delta1 | 0.0410 / 0.9800 | 0.0411 / 0.9800 | 0.0410 / 0.9800 | 0.0410 / 0.9800 | 0.0410 / 0.9800 | |
pointdit-l-512 | iBims-1, AbsRel / delta1 | 0.0313 / 0.9848 | 0.0313 / 0.9847 | 0.0313 / 0.9848 | 0.0313 / 0.9848 | 0.0313 / 0.9848 | |
pointdit-l-512 | ETH3D two-view train left, AbsRel / delta1 | 0.0748 / 0.9264 | 0.0748 / 0.9278 | 0.0748 / 0.9265 | 0.0748 / 0.9261 | possibly in the training data | |
pointdit-l-512 | Middlebury 2014 trainingH left, AbsRel / delta1 | 0.0230 / 0.9952 | 0.0230 / 0.9951 | 0.0230 / 0.9952 | 0.0230 / 0.9952 | possibly in the training data |
Measured on portable: NVIDIA GeForce RTX 5060 Ti, torch 2.14.1+cu130, fastforks 2fb984b3, 2026-11-15; portable32: NVIDIA GeForce RTX 5060 Ti, torch 2.14.1+cu130, fastforks 6a5100ec, 2026-11-15; kernels: NVIDIA GeForce RTX 5060 Ti, torch 2.14.1+cu130, fastforks 2fb984b3, 2026-11-15.
Parity
CPU, float32: max |diff| <= 1.5e-06 vs the original. GPU, float16 (the default), portable and kernels: mean |diff| <= 0.00029 and p99 <= 0.0012 (absolute error of the output points).
| variant | kernels error vs original, mean / p99 | GPU, torch | kernels used |
|---|---|---|---|
pointdit-b-512 | 0.00015 / 0.00056 | RTX 5060 Ti, torch 2.14.1+cu130 | dpt_fused@48e0065bb7cf:sm_120, fastbb@7d058322634d:sm_120 |
pointdit-l-512 | 0.00012 / 0.00045 | RTX 5060 Ti, torch 2.14.1+cu130 | dpt_fused@48e0065bb7cf:sm_120, fastbb@7d058322634d:sm_120 |
Limitations
- Affine-invariant output: points up to an unknown scale and 3-D shift, depth (their z) up to scale and shift; align before use.
- Trained at 512 x 512: inputs are resized to about
tokens16-pixel patches at their aspect ratio (bilinear) and the point map is resized back with nearest, as the original's wild-image path. - Kernels mode captures CUDA graphs: the graph's private memory pool and static outputs stay reserved after the call, so peak reserved is above portable by ~85 MiB (b), more for l (peak allocated is equal or lower);
cuda_graphs=falseorrelease_graphs()frees it at the cost of launch overhead.
Example output
No example rendered for this checkpoint yet.