AutomatosX/AXDocLayout-M-v1.3-CUDA-TF32

AXDocLayout-M v1.3 CUDA TF32

AX DocLayout detects layout regions in a decoded RGB page image and returns class labels, confidence scores and boxes in the original raster coordinates. This model updates the weights while retaining the v1.2 CUDA TF32 inference standard. The predecessor remains available at AXDocLayout-M v1.2 CUDA TF32.

Model and runtime

Architecture: RT-DETRv2 rtdetrv2_r50vd_m. Stored weights and ONNX inputs are FP32. CUDA TF32 is the execution policy, not a BF16 weight format. The bundle contains FP32 safetensors, a standard-operator ONNX graph (opset 16) and the complete inference source. Integer quantization is not used.

The ONNX backend requires Linux x86-64, Python 3.12 or 3.13 and an NVIDIA TF32-capable Ampere or newer device with a compatible CUDA 12.8 driver. Install the pinned ONNX Runtime GPU 1.23.2, CUDA 12.8 and cuDNN 9 dependencies through the onnx extra. CUDAExecutionProvider is required, TF32 is explicitly enabled, and startup checks verify recognized compute placement. Shape/control nodes may use the CPU. An unavailable CUDA backend raises an error.

A separate explicit PyTorch CPU FP32 reference is available with the torch extra and backend="pytorch". Vectorized postprocessing follows v1.2.

Input and output

Input is one HWC uint8 RGB raster. Preserve aspect ratio, resize the long edge to 1024, center-pad to 1024 x 1024 with white, and convert to NCHW float32 in [0, 1]. Contract: rgb-letterbox-1024/1.

Direct ONNX inputs are images float32 [N, 3, 1024, 1024] and orig_target_sizes int64 [N, 2], always [1024, 1024] per image. Outputs are labels, boxes, scores. Raw boxes are letterbox-space xyxy; the Python API undoes padding/scaling and clips boxes to the page. Default score floor is 0.30; invalid or reserved labels are removed.

The JSON schema is ax-doclayout.detection/1; taxonomy is ax-doclayout-11/1. Boxes use half-open pixel edges, origin top-left, x rightward, y downward.

ID	Label
1	caption
2	footnote
3	formula
4	list_item
5	page_footer
6	page_header
7	picture
8	section_header
9	table
10	text
11	title

Qualification and limitations

The following aggregate measurements are internal development results, not independent acceptance or evidence of universal superiority. Both releases were measured on one NVIDIA RTX PRO 6000 Blackwell Server Edition with the same CUDA TF32 runtime. AP averages IoU thresholds 0.50:0.95; AP50, AP70 and AP90 use their respective IoU thresholds. Scores are percentages.

Internal evaluation	Model	AP	AP50	AP70	AP90
A	v1.2	76.2397	90.9894	86.8779	55.9050
A	v1.3	76.1623	90.9487	86.7953	55.8282
B	v1.2	66.7463	78.4563	71.8082	54.7579
B	v1.3	69.4385	80.4662	74.8691	58.1819
C	v1.2	90.8539	95.5602	94.5320	89.6775
C	v1.3	90.9530	95.3348	94.2652	89.3608

Results show mixed tradeoffs, including decreases in some AP and high-IoU metrics. Evaluation groups are anonymous and their definitions are not published; these numbers are not independently reproducible benchmarks and must not be combined with results from other evaluation contexts. Validate on representative inputs before deployment.

The public API latency below measures decoded RGB through PageStructure, excluding raster decoding, JSON serialization and worker transport. It uses the default score floor of 0.30 and batch one on the same host.

Same-host CUDA TF32 API	Median ms	p95 ms	Mean ms
v1.2	17.959	18.268	17.998
v1.3	17.945	18.563	18.013

The ONNX operator structure and numerical inference code follow v1.2; model weights and immutable artifact bindings are updated. The CUDA inference environment does not require PyTorch. GPU execution can vary slightly across invocations; no bitwise repeatability guarantee is made.

The detector accepts RGB page rasters. It does not open PDFs, perform OCR, extract text, predict reading order or hierarchy, or reconstruct table cells.

License

The model weights and modified inference source are AGPL-3.0-only. Commercial use is permitted subject to the license terms, including applicable source-disclosure obligations. See LICENSE and NOTICE. Preserved Apache-2.0 component notices and rights remain in NOTICE. The complete inference source is provided in inference/.

See the root README for immutable source installation and model revisions.

Quick start — CUDA + TF32

Requires supported Linux x86-64/NVIDIA CUDA setup and Python 3.12 or 3.13. Install the download helper, then the immutable inference source:

python -m pip install "huggingface-hub>=0.36,<2"

import subprocess
import sys
from huggingface_hub import snapshot_download

source = snapshot_download(
    "AutomatosX/AXDocLayout-M-v1.3-CUDA-TF32",
    revision="fc27bf145a5e056046ddc94dbe3fc12d45665cc4",
    allow_patterns=["inference/**"],
    token=False,
)
subprocess.check_call([
    sys.executable, "-m", "pip", "install", source + "/inference[onnx]"
])

Run CUDA-powered layout detection:

import numpy as np
from PIL import Image
from ax_doclayout import AXDocLayout, PageImage

model = AXDocLayout.from_pretrained(
    "AutomatosX/AXDocLayout-M-v1.3-CUDA-TF32",
    revision="d2725e733fa857100fe75781715276cee4e5a84f",
    backend="onnx",  # v1.3: CUDAExecutionProvider with TF32 enabled
)
page = PageImage(
    rgb=np.asarray(Image.open("page.png").convert("RGB"), dtype=np.uint8),
    page_index=0,
)
result = model.predict(page)
print(result.to_json())

For the separate CPU PyTorch comparison, install inference[torch] and explicitly select backend="pytorch". The CUDA ONNX loader does not switch to this backend automatically.