libreyolo 1.2.0.dev0__tar.gz → 1.3.1__tar.gz

{libreyolo-1.2.0.dev0 → libreyolo-1.3.1}/MANIFEST.in

@@ -18,9 +18,22 @@ global-exclude *.mp4
 # Re-include packaged sample assets (overrides global-exclude above)
 recursive-include libreyolo/assets *.jpg
 recursive-include libreyolo *.yaml
+recursive-include libreyolo/data *.json
+
+# LibreCLIP shipped data: vendored BPE vocab + ImageNet labels/templates
+recursive-include libreyolo/models/clip *.gz
+recursive-include libreyolo/models/clip/data *.json
+
+# Bundled public-domain Darknet model-definition cfgs (YOLOv2/v3/v4)
+recursive-include libreyolo/models/darknet/cfgs *.cfg
+
+# Bundled MIT YOLOv7 model-definition yaml (MultimediaTechLab/YOLO)
+recursive-include libreyolo/models/yolo7 *.yaml
 
 # Bundled third-party license texts (must ship with the wheel)
 recursive-include libreyolo/models *.md
+recursive-include libreyolo/models NOTICE
 include THIRD_PARTY_NOTICES.txt
 include weights/LICENSE_NOTICE.txt
 include NOTICE
+include licenses/Apache-2.0.txt

libreyolo-1.3.1/NOTICE

@@ -0,0 +1,130 @@
+LibreYOLO
+Copyright (c) The LibreYOLO contributors.
+
+LibreYOLO is licensed under the MIT License (see LICENSE).
+
+This product bundles third-party source code under non-MIT licenses. The
+bundled files retain their original copyright headers. A copy of each
+upstream license accompanies the corresponding code.
+
+------------------------------------------------------------------------
+Bundled third-party source code
+------------------------------------------------------------------------
+
+DINOv3 (Meta DINOv3 License Agreement)
+    Path:    libreyolo/models/deimv2/engine/backbone/dinov3/
+    License: libreyolo/models/deimv2/engine/backbone/dinov3/LICENSE.md
+    Source:  https://github.com/facebookresearch/dinov3
+
+    The DINOv3 backbone code shipped with the DEIMv2 family is
+    redistributed under the DINOv3 License Agreement, a custom
+    non-OSI license from Meta Platforms, Inc.
+
+    Key obligations propagated to downstream users (see LICENSE.md
+    for the authoritative text):
+      * Redistribution must include a copy of the DINOv3 License
+        Agreement (provided as LICENSE.md alongside the code).
+      * The DINO Materials may not be used for activities subject
+        to ITAR, military or warfare purposes, nuclear industries,
+        espionage, or weapons development.
+      * The DINO Materials are provided "as is" without warranty.
+
+    These terms apply only to the DINOv3 subtree listed above. The
+    rest of LibreYOLO remains under the MIT License.
+
+MobileNetV4 / timm (Apache License 2.0)
+    Path:    libreyolo/models/mobilenetv4/
+    License: libreyolo/models/mobilenetv4/NOTICE  (Apache-2.0)
+    Source:  https://github.com/huggingface/pytorch-image-models
+
+    The MobileNetV4 architecture (libreyolo/models/mobilenetv4/nn.py) is a
+    native re-implementation derived from timm's MobileNetV4
+    (Ross Wightman and the timm contributors), licensed Apache-2.0. Module
+    naming mirrors timm so its Apache-2.0 ImageNet-1k pretrained weights load
+    unchanged and inference is bit-identical. Apache-2.0 is MIT-compatible;
+    these terms add only attribution obligations.
+
+ConvNeXt / timm (MIT code + Apache-2.0 weights)
+    Path:    libreyolo/models/convnext/
+    License: libreyolo/models/convnext/NOTICE
+    Source:  https://github.com/facebookresearch/ConvNeXt
+             https://github.com/huggingface/pytorch-image-models
+
+    The ConvNeXt V1 architecture (libreyolo/models/convnext/nn.py) is a native
+    re-implementation derived from Meta's ConvNeXt (MIT) and timm's ConvNeXt.
+    Module naming mirrors timm so its Apache-2.0 ImageNet-1k `fb_in1k` weights
+    load unchanged and inference is bit-identical. ConvNeXt-V2's small
+    checkpoints are CC-BY-NC and are intentionally NOT used.
+
+EfficientNetV2 / timm (Apache License 2.0)
+    Path:    libreyolo/models/efficientnetv2/
+    License: libreyolo/models/efficientnetv2/NOTICE  (Apache-2.0)
+    Source:  https://github.com/huggingface/pytorch-image-models
+             https://github.com/google/automl
+
+    The EfficientNetV2 architecture (libreyolo/models/efficientnetv2/nn.py) is a
+    native re-implementation derived from timm's EfficientNetV2 (Ross Wightman)
+    and Google's EfficientNetV2 (google/automl), licensed Apache-2.0. Module
+    naming mirrors timm so its Apache-2.0 ImageNet-1k weights load unchanged and
+    inference is bit-identical.
+
+ResNet / timm (Apache License 2.0)
+    Path:    libreyolo/models/resnet/
+    License: libreyolo/models/resnet/NOTICE  (Apache-2.0)
+    Source:  https://github.com/huggingface/pytorch-image-models
+
+    The ResNet architecture (libreyolo/models/resnet/nn.py) is a native
+    re-implementation of the vanilla ResNet (He et al. 2015, v1.5). Weights are
+    timm's `resnet*.a1_in1k` (Ross Wightman), Apache-2.0, ImageNet-1k. Module
+    naming mirrors timm/torchvision so weights load unchanged and inference is
+    bit-identical.
+
+NAFNet (MIT License)
+    Path:    libreyolo/models/nafnet/
+    License: libreyolo/models/nafnet/NOTICE  (MIT)
+    Source:  https://github.com/megvii-research/NAFNet
+
+    The NAFNet restoration architecture (libreyolo/models/nafnet/nn.py) is a
+    native PyTorch implementation derived from Megvii Research's NAFNet
+    (MIT). LibreYOLO does not bundle NAFNet pretrained checkpoint files. Some
+    published GoPro-trained NAFNet weights do not carry an explicit standalone
+    weights license; convert only weights that you have the right to use and
+    redistribute.
+
+Darknet YOLOv2 / YOLOv3 / YOLOv4 (public domain)
+    Path:    libreyolo/models/darknet/, libreyolo/models/yolo2|yolo3|yolo4/
+    License: libreyolo/models/darknet/cfgs/NOTICE  (public domain)
+    Source:  https://github.com/pjreddie/darknet  (YOLOv2/v3)
+             https://github.com/AlexeyAB/darknet   (YOLOv4)
+
+    The LibreYOLO2/3/4 families reproduce the YOLOv2, YOLOv3, and YOLOv4
+    architectures from the Darknet project, which is public domain (the
+    "YOLO LICENSE": "Darknet is public domain. Do whatever you want with
+    it."). The model-definition (.cfg) files are bundled under
+    libreyolo/models/darknet/cfgs/ and drive both the runtime graph builder
+    and the weight converter. Only the .cfg format and the numerical
+    behaviour of the Darknet layers are reproduced; no Darknet C source is
+    copied. Being public domain, these terms impose no obligations on
+    LibreYOLO or its downstream users.
+
+YOLOv7 / MultimediaTechLab/YOLO (MIT License)
+    Path:    libreyolo/models/yolo7/
+    Source:  https://github.com/MultimediaTechLab/YOLO
+
+    The LibreYOLO7 family is a native port of YOLOv7 from
+    MultimediaTechLab/YOLO (MIT, (c) 2024 Kin-Yiu Wong & Hao-Tang Tsui) —
+    the authors' own MIT re-release, NOT the GPL-3.0 WongKinYiu/yolov7.
+    Module names mirror upstream so the MIT v7.pt weights load unchanged;
+    the model-definition v7.yaml is bundled under libreyolo/models/yolo7/.
+    MIT is MIT-compatible; these terms add only attribution obligations.
+
+------------------------------------------------------------------------
+Pretrained model weights
+------------------------------------------------------------------------
+
+No pretrained weights are distributed in this source tree. LibreYOLO
+weights are published separately on Hugging Face under the LibreYOLO
+organization (https://huggingface.co/LibreYOLO). Each Hugging Face
+model repository ships its own LICENSE and NOTICE reflecting the
+license of the upstream project the weights were derived from. See
+weights/LICENSE_NOTICE.txt for a per-family summary.

libreyolo-1.3.1/PKG-INFO

@@ -0,0 +1,253 @@
+Metadata-Version: 2.4
+Name: libreyolo
+Version: 1.3.1
+Summary: Libre YOLO - An open source YOLO library with MIT license.
+Author: LibreYOLO Team
+License-Expression: MIT
+Project-URL: Homepage, https://github.com/LibreYOLO
+Project-URL: Repository, https://github.com/LibreYOLO/libreyolo
+Requires-Python: >=3.10
+Description-Content-Type: text/markdown
+License-File: LICENSE
+License-File: NOTICE
+License-File: THIRD_PARTY_NOTICES.txt
+License-File: weights/LICENSE_NOTICE.txt
+License-File: licenses/Apache-2.0.txt
+Requires-Dist: numpy>=1.19.0
+Requires-Dist: Pillow>=9.1.0
+Requires-Dist: torch>=2.4.0
+Requires-Dist: torchvision>=0.19.0
+Requires-Dist: PyYAML>=6.0
+Requires-Dist: requests>=2.25.0
+Requires-Dist: opencv-python>=4.11.0.86
+Requires-Dist: tqdm>=4.65.0
+Requires-Dist: pycocotools>=2.0.0
+Requires-Dist: typer>=0.9.0
+Requires-Dist: click>=8.0.0
+Requires-Dist: safetensors>=0.4.0
+Requires-Dist: scipy>=1.7.0
+Provides-Extra: onnx
+Requires-Dist: onnx>=1.14.0; extra == "onnx"
+Requires-Dist: onnxsim>=0.4.0; extra == "onnx"
+Requires-Dist: onnxruntime>=1.16.0; extra == "onnx"
+Provides-Extra: rfdetr
+Requires-Dist: transformers>=5.1.0; extra == "rfdetr"
+Provides-Extra: eomt
+Requires-Dist: transformers>=5.1.0; extra == "eomt"
+Provides-Extra: lora
+Requires-Dist: libreyolo[rfdetr]; extra == "lora"
+Requires-Dist: peft>=0.12.0; extra == "lora"
+Provides-Extra: plots
+Requires-Dist: matplotlib>=3.7.0; extra == "plots"
+Provides-Extra: vlm
+Requires-Dist: transformers>=5.1.0; extra == "vlm"
+Requires-Dist: num2words>=0.5.0; extra == "vlm"
+Requires-Dist: decord==0.6.0; extra == "vlm"
+Requires-Dist: lmdb>=1.7.0; extra == "vlm"
+Requires-Dist: peft>=0.12.0; extra == "vlm"
+Provides-Extra: sam
+Requires-Dist: transformers>=5.3.0; extra == "sam"
+Provides-Extra: openvocab
+Requires-Dist: transformers>=5.1.0; extra == "openvocab"
+Provides-Extra: label
+Requires-Dist: libreyolo[sam]; extra == "label"
+Provides-Extra: tensorrt
+Requires-Dist: tensorrt-cu12==10.16.1.11; sys_platform != "darwin" and extra == "tensorrt"
+Requires-Dist: pycuda>=2022.1; sys_platform != "darwin" and extra == "tensorrt"
+Provides-Extra: openvino
+Requires-Dist: openvino>=2024.0.0; extra == "openvino"
+Provides-Extra: ncnn
+Requires-Dist: pnnx; extra == "ncnn"
+Requires-Dist: ncnn; extra == "ncnn"
+Provides-Extra: tflite
+Requires-Dist: libreyolo[onnx]; extra == "tflite"
+Requires-Dist: onnx2tf>=2.4.3; python_version >= "3.12" and extra == "tflite"
+Requires-Dist: onnx-graphsurgeon>=0.3.27; extra == "tflite"
+Requires-Dist: onnx-simplifier>=0.4.36; extra == "tflite"
+Provides-Extra: tracking
+Provides-Extra: rtdetr
+Provides-Extra: gaze
+Requires-Dist: gdown>=5.1.0; extra == "gaze"
+Provides-Extra: clip
+Requires-Dist: regex>=2023.0.0; extra == "clip"
+Requires-Dist: ftfy>=6.0.0; extra == "clip"
+Provides-Extra: clip-convert
+Requires-Dist: libreyolo[clip]; extra == "clip-convert"
+Requires-Dist: open_clip_torch>=2.24.0; extra == "clip-convert"
+Provides-Extra: coreml
+Requires-Dist: coremltools>=8.0; extra == "coreml"
+Provides-Extra: tensorboard
+Requires-Dist: tensorboard>=2.13.0; extra == "tensorboard"
+Provides-Extra: mlflow
+Requires-Dist: mlflow>=2.9.0; extra == "mlflow"
+Provides-Extra: wandb
+Requires-Dist: wandb>=0.16.0; extra == "wandb"
+Provides-Extra: all
+Requires-Dist: libreyolo[onnx]; extra == "all"
+Requires-Dist: libreyolo[rfdetr]; extra == "all"
+Requires-Dist: libreyolo[eomt]; extra == "all"
+Requires-Dist: libreyolo[lora]; extra == "all"
+Requires-Dist: libreyolo[plots]; extra == "all"
+Requires-Dist: libreyolo[tensorrt]; extra == "all"
+Requires-Dist: libreyolo[tflite]; extra == "all"
+Requires-Dist: libreyolo[openvino]; extra == "all"
+Requires-Dist: libreyolo[ncnn]; extra == "all"
+Requires-Dist: libreyolo[tracking]; extra == "all"
+Requires-Dist: libreyolo[rtdetr]; extra == "all"
+Requires-Dist: libreyolo[gaze]; extra == "all"
+Requires-Dist: libreyolo[clip]; extra == "all"
+Requires-Dist: libreyolo[coreml]; extra == "all"
+Requires-Dist: libreyolo[vlm]; extra == "all"
+Requires-Dist: libreyolo[sam]; extra == "all"
+Requires-Dist: libreyolo[label]; extra == "all"
+Requires-Dist: libreyolo[openvocab]; extra == "all"
+Requires-Dist: libreyolo[tensorboard]; extra == "all"
+Requires-Dist: libreyolo[mlflow]; extra == "all"
+Requires-Dist: libreyolo[wandb]; extra == "all"
+Dynamic: license-file
+
+# LibreYOLO
+
+[English](README.md) | [简体中文](README.zh-CN.md)
+
+> ⭐ **Support LibreYOLO.** The best way to help is to **star the repo**. Feel free to [open an issue](https://github.com/LibreYOLO/libreyolo/issues/new) if you encounter problems or have suggestions, and code contributions are very welcome (see [CONTRIBUTING.md](CONTRIBUTING.md)).
+
+[![Documentation](https://img.shields.io/badge/docs-libreyolo.com-blue)](https://www.libreyolo.com/docs)
+[![PyPI](https://img.shields.io/pypi/v/libreyolo)](https://pypi.org/project/libreyolo/)
+[![PyPI Downloads](https://static.pepy.tech/badge/libreyolo)](https://pepy.tech/projects/libreyolo)
+[![Hugging Face](https://img.shields.io/badge/%F0%9F%A4%97%20Hugging%20Face-LibreYOLO-yellow)](https://huggingface.co/LibreYOLO)
+[![Benchmarks](https://img.shields.io/badge/benchmarks-visionanalysis.org-purple)](https://www.visionanalysis.org/)
+[![LinkedIn](https://img.shields.io/badge/LinkedIn-LibreYOLO-blue?logo=linkedin)](https://www.linkedin.com/company/libreyolo/)
+[![License](https://img.shields.io/badge/license-MIT-green)](LICENSE)
+
+MIT-licensed computer vision library with inference and training support for a variety of models. It provides a familiar high-level Python and CLI interface and reads common YOLO-format datasets, so existing workflows port over with minimal changes.
+
+![LibreYOLO Detection Example](libreyolo/assets/parkour_result.jpg)
+
+## Installation & Quick start
+
+`pip install libreyolo` covers most users. It comes with the YOLOv9 flagship
+and the other detection models, plus training and inference. Now and then you'll
+add an extra: for a model family with a heavier dependency (for example RF-DETR,
+which needs the large `transformers` library), or for an export backend when you
+need to export a model:
+
+```bash
+pip install libreyolo
+
+# Add an extra in brackets when you need one (comma-separate to combine),
+# e.g. pip install "libreyolo[rfdetr,onnx]":
+#   export:    onnx, tensorrt, openvino, ncnn, tflite, coreml
+#   models:    rfdetr, vlm, sam, openvocab, clip, gaze
+#   training:  lora, plots, tensorboard, mlflow, wandb
+#   or all:    pip install "libreyolo[all]"
+```
+
+```python
+from libreyolo import LibreYOLO, SAMPLE_IMAGE
+
+model = LibreYOLO("LibreYOLO9t.pt")
+result = model(SAMPLE_IMAGE, save=True)
+```
+
+Image classification works the same way. Load a pretrained ImageNet-1k
+classifier (`MobileNetV4`, `ConvNeXt`, `EfficientNetV2`, or `ResNet`), then
+predict or fine-tune on your own folder-per-class dataset:
+
+```python
+from libreyolo import LibreYOLO
+
+model = LibreYOLO("LibreResNet50-cls.pt")   # weights auto-download on first use
+result = model("image.jpg")                  # a single image -> one Results
+print(result.probs.top1, float(result.probs.top1conf))  # class index + confidence
+print(result.probs.top5)                     # indices of the 5 most likely classes
+
+# Fine-tune on an ImageFolder dataset (train/ and val/, one sub-folder per
+# class). The classifier head resizes to your class count automatically.
+model.train(data="path/to/dataset", epochs=5)
+```
+
+For the full list of extras and per-backend notes, see the [docs](https://www.libreyolo.com/docs#installation).
+
+To install from source in editable mode (for development or to track unreleased changes):
+
+```bash
+git clone https://github.com/LibreYOLO/libreyolo.git
+cd libreyolo
+pip install -e .
+```
+
+## Flagship models
+
+LibreYOLO recommends these model families because they offer the best balance
+and receive the heaviest testing:
+
+- **YOLOv9** for CNN-based YOLO models.
+- **RF-DETR** for transformer-based detection and segmentation.
+
+## Compatibility
+
+`✓` supported, `exp` experimental. Empty cells are not currently supported.
+All trainable families in the Training column accept universal training
+hooks via `callbacks=` and built-in experiment loggers via `loggers=`
+(`tensorboard`, `mlflow`, `wandb`).
+<table>
+  <thead>
+    <tr>
+      <th rowspan="2">Model family</th>
+      <th colspan="7">Inference</th>
+      <th rowspan="2">Training</th>
+      <th colspan="6">Export formats</th>
+    </tr>
+    <tr>
+      <th>Detection</th>
+      <th>Segmentation</th>
+      <th>Semantic</th>
+      <th>Classification</th>
+      <th>Pose</th>
+      <th>OBB</th>
+      <th>Gaze</th>
+      <th>ONNX</th>
+      <th>TorchScript</th>
+      <th>TensorRT</th>
+      <th>OpenVINO</th>
+      <th>NCNN</th>
+      <th>TFLite</th>
+    </tr>
+  </thead>
+  <tbody>
+    <tr><td><strong>⭐ YOLOv9</strong></td><td>✓</td><td></td><td></td><td></td><td></td><td></td><td></td><td>✓</td><td>✓</td><td>✓</td><td>✓</td><td>✓</td><td>✓</td><td></td></tr>
+    <tr><td><strong>⭐ RF-DETR</strong></td><td>✓</td><td>✓</td><td>exp</td><td>exp</td><td>exp</td><td>exp</td><td></td><td>✓</td><td>✓</td><td>✓</td><td>✓</td><td>✓</td><td></td><td>exp</td></tr>
+    <tr><td>YOLOX</td><td>✓</td><td></td><td></td><td></td><td></td><td></td><td></td><td>exp</td><td>exp</td><td>exp</td><td>exp</td><td>exp</td><td>exp</td><td></td></tr>
+    <tr><td>YOLOv9-E2E</td><td>✓</td><td></td><td></td><td></td><td></td><td></td><td></td><td>exp</td><td>exp</td><td>exp</td><td>exp</td><td></td><td></td><td></td></tr>
+    <tr><td>YOLOv9-P2</td><td>✓</td><td></td><td></td><td></td><td></td><td></td><td></td><td>✓</td><td>exp</td><td></td><td></td><td></td><td></td><td></td></tr>
+    <tr><td>YOLO-NAS</td><td>✓</td><td></td><td></td><td></td><td>✓</td><td></td><td></td><td>exp</td><td>exp</td><td>exp</td><td>exp</td><td>exp</td><td>exp</td><td></td></tr>
+    <tr><td>D-FINE</td><td>✓</td><td></td><td></td><td></td><td></td><td></td><td></td><td>exp</td><td>exp</td><td>exp</td><td>exp</td><td>exp</td><td></td><td></td></tr>
+    <tr><td>DEIM</td><td>✓</td><td></td><td></td><td></td><td></td><td></td><td></td><td>exp</td><td>exp</td><td>exp</td><td>exp</td><td>exp</td><td></td><td></td></tr>
+    <tr><td>DEIMv2</td><td>✓</td><td></td><td></td><td></td><td></td><td></td><td></td><td>exp</td><td>exp</td><td>exp</td><td>exp</td><td>exp</td><td></td><td></td></tr>
+    <tr><td>RT-DETR</td><td>✓</td><td></td><td></td><td></td><td></td><td></td><td></td><td>exp</td><td>exp</td><td>exp</td><td>exp</td><td>exp</td><td></td><td></td></tr>
+    <tr><td>RT-DETRv2</td><td>✓</td><td></td><td></td><td></td><td></td><td></td><td></td><td>exp</td><td></td><td></td><td></td><td></td><td></td><td></td></tr>
+    <tr><td>RT-DETRv4</td><td>✓</td><td></td><td></td><td></td><td></td><td></td><td></td><td>exp</td><td></td><td></td><td></td><td></td><td></td><td></td></tr>
+    <tr><td>PicoDet</td><td>✓</td><td></td><td></td><td></td><td></td><td></td><td></td><td>exp</td><td>exp</td><td>exp</td><td></td><td></td><td></td><td></td></tr>
+    <tr><td>RTMDet</td><td>✓</td><td></td><td></td><td></td><td></td><td></td><td></td><td>exp</td><td></td><td></td><td></td><td></td><td></td><td></td></tr>
+    <tr><td>EC</td><td>✓</td><td>✓</td><td></td><td></td><td>✓</td><td></td><td></td><td>exp</td><td></td><td></td><td></td><td></td><td></td><td></td></tr>
+    <tr><td>MobileNetV4</td><td></td><td></td><td></td><td>✓</td><td></td><td></td><td></td><td>✓</td><td>✓</td><td></td><td></td><td></td><td></td><td></td></tr>
+    <tr><td>ConvNeXt</td><td></td><td></td><td></td><td>✓</td><td></td><td></td><td></td><td>✓</td><td>✓</td><td></td><td></td><td></td><td></td><td></td></tr>
+    <tr><td>EfficientNetV2</td><td></td><td></td><td></td><td>✓</td><td></td><td></td><td></td><td>✓</td><td>✓</td><td></td><td></td><td></td><td></td><td></td></tr>
+    <tr><td>ResNet</td><td></td><td></td><td></td><td>✓</td><td></td><td></td><td></td><td>✓</td><td>✓</td><td></td><td></td><td></td><td></td><td></td></tr>
+    <tr><td>L2CS</td><td></td><td></td><td></td><td></td><td></td><td></td><td>✓</td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr>
+  </tbody>
+</table>
+
+YOLOv9-P2 is a small-object variant of YOLOv9 with an extra stride-4 detection
+scale, built for aerial/tiny-object imagery where objects fall below ~16 px
+(on regular datasets like COCO, prefer stock YOLOv9). A VisDrone-trained
+research preview is available as
+[`LibreYOLO9P2s-visdrone.pt`](https://huggingface.co/LibreYOLO/LibreYOLO9P2s-visdrone)
+(non-commercial license); train your own with
+`LibreYOLO9P2(None, size="s").train(..., pretrained="LibreYOLO9s.pt")`.
+
+## License
+
+- **Code:** MIT License
+- **Weights:** Pre-trained weights may inherit licensing from the original source. Check the license in the specific HF repo of weights that you are interested in. LibreYOLO HF models always have a license.

libreyolo-1.3.1/README.md

@@ -0,0 +1,145 @@
+# LibreYOLO
+
+[English](README.md) | [简体中文](README.zh-CN.md)
+
+> ⭐ **Support LibreYOLO.** The best way to help is to **star the repo**. Feel free to [open an issue](https://github.com/LibreYOLO/libreyolo/issues/new) if you encounter problems or have suggestions, and code contributions are very welcome (see [CONTRIBUTING.md](CONTRIBUTING.md)).
+
+[![Documentation](https://img.shields.io/badge/docs-libreyolo.com-blue)](https://www.libreyolo.com/docs)
+[![PyPI](https://img.shields.io/pypi/v/libreyolo)](https://pypi.org/project/libreyolo/)
+[![PyPI Downloads](https://static.pepy.tech/badge/libreyolo)](https://pepy.tech/projects/libreyolo)
+[![Hugging Face](https://img.shields.io/badge/%F0%9F%A4%97%20Hugging%20Face-LibreYOLO-yellow)](https://huggingface.co/LibreYOLO)
+[![Benchmarks](https://img.shields.io/badge/benchmarks-visionanalysis.org-purple)](https://www.visionanalysis.org/)
+[![LinkedIn](https://img.shields.io/badge/LinkedIn-LibreYOLO-blue?logo=linkedin)](https://www.linkedin.com/company/libreyolo/)
+[![License](https://img.shields.io/badge/license-MIT-green)](LICENSE)
+
+MIT-licensed computer vision library with inference and training support for a variety of models. It provides a familiar high-level Python and CLI interface and reads common YOLO-format datasets, so existing workflows port over with minimal changes.
+
+![LibreYOLO Detection Example](libreyolo/assets/parkour_result.jpg)
+
+## Installation & Quick start
+
+`pip install libreyolo` covers most users. It comes with the YOLOv9 flagship
+and the other detection models, plus training and inference. Now and then you'll
+add an extra: for a model family with a heavier dependency (for example RF-DETR,
+which needs the large `transformers` library), or for an export backend when you
+need to export a model:
+
+```bash
+pip install libreyolo
+
+# Add an extra in brackets when you need one (comma-separate to combine),
+# e.g. pip install "libreyolo[rfdetr,onnx]":
+#   export:    onnx, tensorrt, openvino, ncnn, tflite, coreml
+#   models:    rfdetr, vlm, sam, openvocab, clip, gaze
+#   training:  lora, plots, tensorboard, mlflow, wandb
+#   or all:    pip install "libreyolo[all]"
+```
+
+```python
+from libreyolo import LibreYOLO, SAMPLE_IMAGE
+
+model = LibreYOLO("LibreYOLO9t.pt")
+result = model(SAMPLE_IMAGE, save=True)
+```
+
+Image classification works the same way. Load a pretrained ImageNet-1k
+classifier (`MobileNetV4`, `ConvNeXt`, `EfficientNetV2`, or `ResNet`), then
+predict or fine-tune on your own folder-per-class dataset:
+
+```python
+from libreyolo import LibreYOLO
+
+model = LibreYOLO("LibreResNet50-cls.pt")   # weights auto-download on first use
+result = model("image.jpg")                  # a single image -> one Results
+print(result.probs.top1, float(result.probs.top1conf))  # class index + confidence
+print(result.probs.top5)                     # indices of the 5 most likely classes
+
+# Fine-tune on an ImageFolder dataset (train/ and val/, one sub-folder per
+# class). The classifier head resizes to your class count automatically.
+model.train(data="path/to/dataset", epochs=5)
+```
+
+For the full list of extras and per-backend notes, see the [docs](https://www.libreyolo.com/docs#installation).
+
+To install from source in editable mode (for development or to track unreleased changes):
+
+```bash
+git clone https://github.com/LibreYOLO/libreyolo.git
+cd libreyolo
+pip install -e .
+```
+
+## Flagship models
+
+LibreYOLO recommends these model families because they offer the best balance
+and receive the heaviest testing:
+
+- **YOLOv9** for CNN-based YOLO models.
+- **RF-DETR** for transformer-based detection and segmentation.
+
+## Compatibility
+
+`✓` supported, `exp` experimental. Empty cells are not currently supported.
+All trainable families in the Training column accept universal training
+hooks via `callbacks=` and built-in experiment loggers via `loggers=`
+(`tensorboard`, `mlflow`, `wandb`).
+<table>
+  <thead>
+    <tr>
+      <th rowspan="2">Model family</th>
+      <th colspan="7">Inference</th>
+      <th rowspan="2">Training</th>
+      <th colspan="6">Export formats</th>
+    </tr>
+    <tr>
+      <th>Detection</th>
+      <th>Segmentation</th>
+      <th>Semantic</th>
+      <th>Classification</th>
+      <th>Pose</th>
+      <th>OBB</th>
+      <th>Gaze</th>
+      <th>ONNX</th>
+      <th>TorchScript</th>
+      <th>TensorRT</th>
+      <th>OpenVINO</th>
+      <th>NCNN</th>
+      <th>TFLite</th>
+    </tr>
+  </thead>
+  <tbody>
+    <tr><td><strong>⭐ YOLOv9</strong></td><td>✓</td><td></td><td></td><td></td><td></td><td></td><td></td><td>✓</td><td>✓</td><td>✓</td><td>✓</td><td>✓</td><td>✓</td><td></td></tr>
+    <tr><td><strong>⭐ RF-DETR</strong></td><td>✓</td><td>✓</td><td>exp</td><td>exp</td><td>exp</td><td>exp</td><td></td><td>✓</td><td>✓</td><td>✓</td><td>✓</td><td>✓</td><td></td><td>exp</td></tr>
+    <tr><td>YOLOX</td><td>✓</td><td></td><td></td><td></td><td></td><td></td><td></td><td>exp</td><td>exp</td><td>exp</td><td>exp</td><td>exp</td><td>exp</td><td></td></tr>
+    <tr><td>YOLOv9-E2E</td><td>✓</td><td></td><td></td><td></td><td></td><td></td><td></td><td>exp</td><td>exp</td><td>exp</td><td>exp</td><td></td><td></td><td></td></tr>
+    <tr><td>YOLOv9-P2</td><td>✓</td><td></td><td></td><td></td><td></td><td></td><td></td><td>✓</td><td>exp</td><td></td><td></td><td></td><td></td><td></td></tr>
+    <tr><td>YOLO-NAS</td><td>✓</td><td></td><td></td><td></td><td>✓</td><td></td><td></td><td>exp</td><td>exp</td><td>exp</td><td>exp</td><td>exp</td><td>exp</td><td></td></tr>
+    <tr><td>D-FINE</td><td>✓</td><td></td><td></td><td></td><td></td><td></td><td></td><td>exp</td><td>exp</td><td>exp</td><td>exp</td><td>exp</td><td></td><td></td></tr>
+    <tr><td>DEIM</td><td>✓</td><td></td><td></td><td></td><td></td><td></td><td></td><td>exp</td><td>exp</td><td>exp</td><td>exp</td><td>exp</td><td></td><td></td></tr>
+    <tr><td>DEIMv2</td><td>✓</td><td></td><td></td><td></td><td></td><td></td><td></td><td>exp</td><td>exp</td><td>exp</td><td>exp</td><td>exp</td><td></td><td></td></tr>
+    <tr><td>RT-DETR</td><td>✓</td><td></td><td></td><td></td><td></td><td></td><td></td><td>exp</td><td>exp</td><td>exp</td><td>exp</td><td>exp</td><td></td><td></td></tr>
+    <tr><td>RT-DETRv2</td><td>✓</td><td></td><td></td><td></td><td></td><td></td><td></td><td>exp</td><td></td><td></td><td></td><td></td><td></td><td></td></tr>
+    <tr><td>RT-DETRv4</td><td>✓</td><td></td><td></td><td></td><td></td><td></td><td></td><td>exp</td><td></td><td></td><td></td><td></td><td></td><td></td></tr>
+    <tr><td>PicoDet</td><td>✓</td><td></td><td></td><td></td><td></td><td></td><td></td><td>exp</td><td>exp</td><td>exp</td><td></td><td></td><td></td><td></td></tr>
+    <tr><td>RTMDet</td><td>✓</td><td></td><td></td><td></td><td></td><td></td><td></td><td>exp</td><td></td><td></td><td></td><td></td><td></td><td></td></tr>
+    <tr><td>EC</td><td>✓</td><td>✓</td><td></td><td></td><td>✓</td><td></td><td></td><td>exp</td><td></td><td></td><td></td><td></td><td></td><td></td></tr>
+    <tr><td>MobileNetV4</td><td></td><td></td><td></td><td>✓</td><td></td><td></td><td></td><td>✓</td><td>✓</td><td></td><td></td><td></td><td></td><td></td></tr>
+    <tr><td>ConvNeXt</td><td></td><td></td><td></td><td>✓</td><td></td><td></td><td></td><td>✓</td><td>✓</td><td></td><td></td><td></td><td></td><td></td></tr>
+    <tr><td>EfficientNetV2</td><td></td><td></td><td></td><td>✓</td><td></td><td></td><td></td><td>✓</td><td>✓</td><td></td><td></td><td></td><td></td><td></td></tr>
+    <tr><td>ResNet</td><td></td><td></td><td></td><td>✓</td><td></td><td></td><td></td><td>✓</td><td>✓</td><td></td><td></td><td></td><td></td><td></td></tr>
+    <tr><td>L2CS</td><td></td><td></td><td></td><td></td><td></td><td></td><td>✓</td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr>
+  </tbody>
+</table>
+
+YOLOv9-P2 is a small-object variant of YOLOv9 with an extra stride-4 detection
+scale, built for aerial/tiny-object imagery where objects fall below ~16 px
+(on regular datasets like COCO, prefer stock YOLOv9). A VisDrone-trained
+research preview is available as
+[`LibreYOLO9P2s-visdrone.pt`](https://huggingface.co/LibreYOLO/LibreYOLO9P2s-visdrone)
+(non-commercial license); train your own with
+`LibreYOLO9P2(None, size="s").train(..., pretrained="LibreYOLO9s.pt")`.
+
+## License
+
+- **Code:** MIT License
+- **Weights:** Pre-trained weights may inherit licensing from the original source. Check the license in the specific HF repo of weights that you are interested in. LibreYOLO HF models always have a license.