AutoVision
企业级视觉模型训练框架:自监督预训练 / 蒸馏 → 微调 → 评测 → 导出一条链走完。
覆盖检测、旋转框、实例 / 全景 / 语义分割、姿态、深度、分类;Python API 与
autovision-cli 共享同一套 Pydantic 严格校验。
发行名说明:PyPI 包名是
autovision-framework,import 包名仍是autovision,命令行仍是autovision-cli。本项目与 PyPI 上无关的autovision0.0.1 包没有关系;请确认安装的是autovision-framework。
安装
Python ≥ 3.10:
pip install autovision-framework
# 常用组合:HF 骨干 + timm + ONNX 导出
pip install "autovision-framework[transformers,timm,export]"
# 预训练 RF-DETR 直接推理
pip install "autovision-framework[rfdetr,export]"
模型包按 extras 惰性加载,缺失时对应清单为空而不是报错。完整 extras 表见 仓库 README。
快速开始
不训练,直接加载发布权重并预测:
import autovision
model = autovision.detection("rfdetr/rf-detr-small", pretrained=True)
result = model.predict("image.jpg") # {'bboxes': ..., 'scores': ..., 'labels': ...}
model.export("model.onnx") # parity 门禁 + postprocess 规格
目标检测,训练 → 验证 → 预测 → 导出:
import autovision
data = autovision.ensure_coco8() # 合成冒烟数据集;可换 COCO/YOLO 配置
model = autovision.detection("rtdetr/resnet18").train(
out="runs/det1", data=data, steps=2000, batch_size="auto",
)
model.val(data=data) # {'map': ..., 'map_50': ...}
model.predict("image.jpg") # {'bboxes': ..., 'scores': ..., 'labels': ...}
model.export("model.onnx") # 按 .onnx / .engine / .pt 后缀分发
等价命令行(同一套 Pydantic 校验):
autovision-cli train-object-detection out=runs/det1 data=data.yaml steps=2000
autovision-cli val model=runs/det1/exported_models/exported_best.pt data=coco8
autovision-cli predict model=runs/det1/exported_models/exported_best.pt images=dir/ out=preds/ threshold=0.5
常用 extras
| extra | 能力 |
|---|---|
transformers |
DINOv2 / DINOv3 骨干(HuggingFace 权重) |
timm |
timm 全量骨干(任意输入通道数) |
export |
ONNX 导出 + 数值对拍门禁 |
openvino |
OpenVINO IR 导出与运行时 |
video |
视频 / RTSP 参考推理运行时 |
serving |
FastAPI 参考 serving(/predict、/metrics) |
gtr |
GTR GLA 骨干(Triton,需 CUDA) |
download |
GTR 上游权重按需下载(huggingface_hub) |
dev |
pytest / ruff(CI 同款) |
许可
Apache-2.0(AutoVision 自有代码;全文:
https://www.apache.org/licenses/LICENSE-2.0)。第三方组件与权重溯源见发行包内的
NOTICE 与 THIRD_PARTY_LICENSES.md。受限许可权重(DINOv2 CC-BY-NC 4.0 / DINOv3
License)默认不下载;用户自持权重不可再分发。
0.3.0 许可状态:DEIMv2 派生的
SpatialPriorModule已删除;DINOv3 ViT 变体保留并使用 no-STA adapter,发行物不再包含 DEIMv2 代码。旧的backbone.sta.*ViT 检查点不受支持,需重新训练/导出。详见docs/provenance/LEDGER.md。
完整文档(模型库、基准、导出细节、已知限制)见
文档站 与源码仓库的 docs/ 目录。
Metadata
Release files for autovision-framework 0.3.0
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | Provenance | Actions |
|---|---|---|---|---|
| autovision_framework-0.3.0.tar.gz | 1.5 MB | With provenance | Details for autovision_framework-0.3.0.tar.gz |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Actions Reset |
|---|---|---|---|---|
| autovision_framework-0.3.0-py3-none-any.whl | Python 3 | none | any | Details for autovision_framework-0.3.0-py3-none-any.whl |
Total release size: 2.6 MB
Release files / autovision_framework-0.3.0.tar.gz
| Download URL | autovision_framework-0.3.0.tar.gz |
|---|---|
| Size | 1.5 MB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
417c179b71ebfe4751f2c422ae5bb3e9d8c2fbe7a28a398a3f70b251a683d85e
|
|
BLAKE2b-256 checksum How to use checksums |
9ce28ef7535ac76dceb91ef73c339370f1b9b5438056fc23625ff948184cd0ab
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/7.0.0 CPython/3.13.14
|
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Oct 19, 2026.
Transparency logRelease files / autovision_framework-0.3.0-py3-none-any.whl
| Download URL | autovision_framework-0.3.0-py3-none-any.whl |
|---|---|
| Size | 1.1 MB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
31b1ce894e4702d33f163e1bf0712f7419f14f17589a1aff31be529d47e73a3a
|
|
BLAKE2b-256 checksum How to use checksums |
1ebd1aa9351e45a8cba938c9ca8068eea47f3187d94f4e78d04255bfda9d0751
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/7.0.0 CPython/3.13.14
|
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Oct 19, 2026.
Transparency logSecurity for autovision-framework 0.3.0
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
Full provenance
Both files in this release have been uploaded with provenance metadata.
Attestation for autovision_framework-0.3.0.tar.gz
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Oct 19, 2026.
Transparency logA note on package security
PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.
To learn more about protecting yourself from malicious actors, visit our package security documentation.