MAAP3D/dependency-wheels

MAAP3D patched dependency wheels

This is a versioned artifact repository, not a dataset or a Python package index. It publishes pinned downstream wheels used to beta-test specific MAAP3D deployment profiles. These artifacts are not official releases of the upstream projects.

The wheels contain third-party code and retain their applicable upstream licenses. They are not relicensed under MAAP3D's AGPL license. See LICENSE, the individual files under licenses/, and THIRD_PARTY_WHEEL_NOTICES.zip before redistributing or using an artifact.

Published artifacts

Distribution	Downstream version	Intended MAAP3D profile	License
mmcv	2.2.0, Windows cp310	Torch 2.4.1/cu121; SM75/80/86/89	Apache-2.0 generally; see MMCV special-operation NVIDIA notices
mmcv	2.2.0, build 1cu128sm120, Windows cp310	Experimental Torch 2.7.1/cu128; SM120 only	Apache-2.0 generally; see MMCV special-operation NVIDIA notices
mmcv	2.2.0, build 1cpupt241, Windows cp310	Experimental Torch 2.4.1+cpu; AMD64	Apache-2.0 generally; see MMCV operation-specific notices
mmcv	2.2.0, build 1cpupt240, Linux cp310	Torch 2.4.0+cpu; x86-64 / glibc >=2.17	Apache-2.0 generally; see MMCV operation-specific notices
mmdet	3.3.0+maap3d.cfd5d3a9.1	CUDA 12.1 and CUDA 12.8 full/dev	Apache-2.0
sam3	0.1.0+maap3d.847e1a3b.1	Legacy CUDA 12.8 artifact with PyTorch >=2.7	Custom SAM License dated November 19, 2025
sam3	0.1.0+maap3d.847e1a3b.2	CUDA 12.1 and CUDA 12.8 full/dev with PyTorch >=2.4	Custom SAM License dated November 19, 2025
chumpy	0.71+maap3d.580566ea.1	MAAP3D OpenMMLab stacks, including Python 3.12 GB10	MIT
mmpose	1.3.2+maap3d.gb10.5408bc76.1	GB10 mmcv-lite only	Apache-2.0 generally; EDPose files are identified as IDEA License 1.0

This repository contains pure-Python wheels and compiled Windows and Linux CPU MMCV wheels. Each artifact is specific to its stated Python, PyTorch, CUDA, and platform profile. Do not install every artifact together as a generic environment.

Experimental Windows CUDA 12.8 / SM120 MMCV wheel

mmcv-2.2.0-1cu128sm120-cp310-cp310-win_amd64.whl is built from unmodified MMCV 2.2.0 source for native Windows, 64-bit CPython 3.10, PyTorch 2.7.1+cu128, and compute capability 12.0 only. It contains 58 CUDA ELF binaries targeting SM120. It must not replace the separate CUDA 12.1 wheel or be used on Linux, ARM64, Python 3.11, or other Blackwell compute capabilities.

A fresh Windows RTX 5050 Laptop GPU environment passed a synchronized Torch matrix check, MMCV CUDA NMS, RoIAlign forward/backward against Torchvision, deformable convolution forward/backward against ordinary convolution, and modulated deformable convolution forward/backward. Its pip check passed. Publication checks verified the exact SHA-256, ZIP/RECORD integrity, AMD64 PE header, and embedded CUDA architecture flags. Reports and reproducible build scripts are under builds/windows-cu128-sm120/.

Full MAAP3D validation has NOT RUN: the full installer, MMPose, MMDetection, SAM3, and real checkpoint/video workflows remain pending. This is an experimental component artifact, not a validated full environment. The CUDA 12.8 MAAP3D installer downloads it with an immutable revision and SHA-256 pin, installs it with --no-deps, and retains the full runtime checks. It requires every visible NVIDIA GPU to have compute capability 12.0.

No compiler or full CUDA toolkit is needed to install this prebuilt MMCV wheel. Its DLL imports require the matching Python/PyTorch CUDA runtime and Microsoft Visual C++ 2015–2022 x64 runtime (MSVCP140/VCRUNTIME140). SAM3 may have additional runtime compilation requirements.

MMCV's LICENSES.md identifies NVIDIA-licensed special operations. Complete StyleGAN2 and StyleGAN3 license texts accompany these artifacts; their research/evaluation use restrictions remain applicable. The general Apache license does not override those operation-specific terms.

Experimental Windows CPU MMCV wheel

mmcv-2.2.0-1cpupt241-cp310-cp310-win_amd64.whl uses unmodified MMCV 2.2.0 source, CPU operators only, native Windows AMD64, CPython 3.10, and exactly Torch 2.4.1+cpu (TorchVision 0.19.1+cpu / TorchAudio 2.4.1+cpu). Python 3.11, Windows ARM64 and CUDA Torch are outside this artifact's ABI. Wheel tags do not encode the Torch ABI. SHA-256: 35db4db7d467f45e71e4d50c19ffaf6e35c713053109ff413d199ac503a45389.

The Windows build used VS 2026/MSVC v145 (19.51.36260), an approved deviation from the handoff's VS 2022/v143 target, with no source patches. Supplied build and separate-environment logs show CPU NMS, RoIAlign forward/backward and pip check passing. Independent Linux publication checks verified the exact SHA-256, all 537 RECORD hashes/sizes, ZIP integrity, AMD64 PE extension, licenses, and DLL imports with no CUDA dependencies. Original scripts and reports are preserved under builds/windows-cpu/. Historical candidate-only wording in those original reports describes their build-time status; this section records the subsequent publication.

python install_package.py --cpu in the updated MAAP3D checkout selects this wheel from an immutable revision, verifies SHA-256, and force-reinstalls with --no-deps. No GPU, CUDA toolkit or MMCV compiler is required. Install the Microsoft Visual C++ x64 Redistributable for the MSVCP140/VCRUNTIME140 runtime DLLs, alongside matching CPU Torch/Python.

Full Windows MAAP3D installation, OpenMMLab model checks, checkpoint-backed inference, and a clean Windows machine without Build Tools have NOT RUN. The installer retains its CPU operator/model smoke and final pip check. This remains an experimental component release pending those checks.

Linux CPU MMCV wheel

mmcv-2.2.0-1cpupt240-cp310-cp310-manylinux2014_x86_64.manylinux_2_17_x86_64.whl is built from unmodified MMCV 2.2.0 for CPython 3.10, Linux/WSL2 x86-64, glibc >=2.17 and exactly Torch 2.4.0+cpu (TorchVision 0.19.0+cpu / TorchAudio 2.4.0+cpu). SHA-256: d53f33353426cc2e17a170956c86c7920356b513b1eb4cf1dc80165096f0b970.

This wheel contains CPU operators and no CUDA dependencies. Matching Torch libraries remain external: the manylinux tags do not make it a standalone manylinux distribution or encode the Torch ABI. Python 3.11, ARM64, 32-bit Python, musl/Alpine, and CUDA Torch are outside this artifact's target. Import Torch before directly loading MMCV operators.

The manylinux2014/glibc 2.17 build and fresh verification environment passed CPU NMS, RoIAlign forward/backward, checkpoint-free MMPose HRNet and MMDetection Mask R-CNN feature checks, and pip check. Independent publication review checked SHA-256, ZIP CRC, 537 RECORD hashes/sizes, ELF64 x86-64 identity, licenses and all shared-library dependencies. Those runtime smokes and pip check also passed on the Linux publication host using the exact wheel installed into a temporary target over the existing CPU runtime. Build scripts, original logs and publication checks are under builds/linux-cpu/. Original candidate-only report wording records build-time status before this publication.

The updated MAAP3D python install_package.py --cpu selects the wheel at an immutable revision, verifies SHA-256, and force-reinstalls with --no-deps. It checks Python, architecture and glibc compatibility before changing packages. No MMCV source compilation, CUDA toolkit or NVIDIA GPU is required. Complete installation on a clean host without build tools and real checkpoint-backed inference remain NOT RUN.

What was changed

• MMDetection: widens the package's runtime MMCV version guard so the validated MAAP3D stack can use MMCV 2.2.0, adds an inline downstream modification notice, and uses a downstream version. The optional mim dependency metadata retains upstream's mmcv<2.2.0 declaration; this wheel is therefore not intended to be installed through the [mim] extra.
• SAM3: relaxes the ftfy dependency for the BoxMOT-compatible stack, declares an explicit PyTorch minimum, removes an incorrect MIT classifier, adds a modification notice, and uses a downstream version. Version .2 lowers the declared minimum from PyTorch 2.7 to the validated PyTorch 2.4 runtime without replacing the immutable .1 artifact. The SAM License is a custom license with use and redistribution conditions; read licenses/sam3/LICENSE in full.
• Chumpy: packages the pinned upstream Python 3.12-compatible revision instead of patching Chumpy 0.70 in site-packages, and uses a downstream version.
• MMPose: omits eager EDPose and RTMO registry imports needed by unsupported GB10 paths and uses a downstream version. EDPose source and configuration files remain in the wheel; the special-file licensing recorded in licenses/mmpose_gb10/LICENSES.md still applies. Consult that notice together with the original ED-Pose license file; this repository does not resolve or simplify those upstream terms.

Exact upstream repositories, commits, checksums, and modification summaries are recorded in PROVENANCE.json.

Installation and profile restrictions

The preferred installation path is the matching MAAP3D profile installer, which is responsible for selecting compatible PyTorch, CUDA, MMCV, and other dependencies. These wheels are not complete environment bundles.

• The GB10 MMPose wheel must be installed with --no-deps after the profile-specific xtcocotools build.
• Use SAM3 .2 with the modern CUDA 12.1 / PyTorch 2.4 profile. The older .1 wheel retains its original PyTorch >=2.7 metadata for reproducibility.
• The MMPose wheel is limited to the GB10 mmcv-lite profile. Its disabled eager registrations mean the EDPose and MMDetection-dependent RTMO paths are not supported in that profile.
• The MMDetection wheel changes the runtime compatibility guard only; do not request its optional [mim] extra.

For a manual download, select one artifact and pin a published tag or, for maximum immutability, the full repository commit SHA:

hf download MAAP3D/dependency-wheels \
  FILENAME.whl \
  --repo-type dataset \
  --revision TAG_OR_FULL_COMMIT_SHA \
  --local-dir ./maap3d-wheels

Install the downloaded wheel only after preparing the matching profile. For example, the GB10 MMPose artifact deliberately requires --no-deps:

python -m pip install --no-deps ./maap3d-wheels/mmpose-1.3.2+maap3d.gb10.5408bc76.1-py3-none-any.whl

Integrity and provenance

After downloading the complete artifact set, verify it with:

sha256sum -c SHA256SUMS

SHA256SUMS covers the nine wheels, the original THIRD_PARTY_WHEEL_NOTICES.zip, and the additive THIRD_PARTY_WHEEL_NOTICES-windows-cpu-20261025.zip and THIRD_PARTY_WHEEL_NOTICES-linux-cpu-20261025.zip. It does not cover repository documentation, the standalone license copies, or PROVENANCE.json.

The original notices archive retains its earlier release snapshot. The additive Windows CPU notices archive contains the updated release notice/provenance and applicable MMCV license copies, without replacing the existing archive. The additive Linux CPU notices archive contains the subsequent release notice/provenance and applicable MMCV licenses. Each wheel also carries its applicable package license metadata. Model checkpoints, PyTorch/CUDA runtimes, xtcocotools, and other transitive dependencies are not included. Model checkpoints can have terms separate from their source code and must be obtained from an authorized source.

Release maintenance

Published artifacts must be treated as immutable:

1. Never replace a wheel while retaining its filename or downstream version.
2. Rebuild changed artifacts under a new downstream version.
3. Regenerate SHA256SUMS, PROVENANCE.json, and the notices archive.
4. Commit the complete set and create a new repository tag.
5. Pin automated installers to a full commit SHA when strict reproducibility is required; tags are convenient release labels but can technically be moved.

This repository currently contains beta-test artifacts for the MAAP3D 0.5 development line.