easydetect 0.2.0__tar.gz → 0.2.1__tar.gz

{easydetect-0.2.0 → easydetect-0.2.1}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.5
 Name: easydetect
-Version: 0.2.0
+Version: 0.2.1
 Summary: Easy real-time object detection you can ship: D-FINE with PyTorch training, OpenVINO inference on CPU/GPU/NPU, and a browser app to label and train. Apache-2.0 throughout.
 Project-URL: Homepage, https://github.com/themakerrobot/easydetect
 Project-URL: Source, https://github.com/themakerrobot/easydetect
@@ -145,18 +145,21 @@ faster on a small set, and `resume=True` picks a killed run back up.
 
 ## Label, train and watch it in a browser
 
-`platform/` is a web app on top of this package: drop images in, label them (the
-model drafts the boxes, you correct them), queue a training run, watch the curve,
-then run the result over a folder, a video or your webcam — on one machine, with
-nothing leaving it. A finished run hands you copy-ready code and a Hugging Face
-folder whose model card is written from the run.
+[easydetect lab](https://github.com/themakerrobot/easydetect-lab) is a web app on top of this package, in its own
+repository: drop images in, label them (the model drafts the boxes, you correct
+them), queue a training run, watch the curve, then run the result over a folder,
+a video or your webcam — on one machine, with nothing leaving it. A finished run
+hands you copy-ready code and a Hugging Face folder whose model card is written
+from the run.
 
 ```bash
-pip install -r platform/requirements.txt
-python platform/run.py          # http://<this machine>:8080
+git clone https://github.com/themakerrobot/easydetect-lab
+cd easydetect-lab
+pip install -r requirements.txt     # easydetect[train] from PyPI, and the web server
+python run.py                       # http://<this machine>:8080
 ```
 
-![the platform](https://raw.githubusercontent.com/themakerrobot/easydetect/main/docs/assets/platform.jpg)
+![easydetect lab](https://raw.githubusercontent.com/themakerrobot/easydetect-lab/main/docs/lab.jpg)
 
 ## Models
 
@@ -226,7 +229,7 @@ easydetect export  model=best.pt format=openvino half=true
 
 * [Using the model](https://github.com/themakerrobot/easydetect/blob/main/docs/usage.md) — sources, results, tracking, saving
 * [Training](https://github.com/themakerrobot/easydetect/blob/main/docs/training.md) — datasets, validation, export, CLI
-* [Platform](https://github.com/themakerrobot/easydetect/blob/main/platform/README.md) — the browser app: labelling, jobs, sharing
+* [easydetect lab](https://github.com/themakerrobot/easydetect-lab) — the browser app: labelling, jobs, sharing
 * [Weights](https://github.com/themakerrobot/easydetect/blob/main/docs/weights.md) — the mirror, building it, offline use
 * [Performance](https://github.com/themakerrobot/easydetect/blob/main/docs/performance.md) — measuring speed and improving it
 * [Design](https://github.com/themakerrobot/easydetect/blob/main/docs/design.md) — architecture and provenance
@@ -259,7 +262,7 @@ Runtime이 함께 들어가고 PyTorch는 없습니다. `pip install "easydetect
 `Detector("dfine-s", backend="onnxruntime")`로 ONNX Runtime(모든 CPU, 라즈베리파이
 포함)을 쓸 수 있습니다. 두 엔진의 결과는 같습니다.
 
-라벨링부터 학습·추론까지 브라우저로 하려면 `python platform/run.py`. 가중치 캐시는
+라벨링부터 학습·추론까지 브라우저로 하려면 [easydetect lab](https://github.com/themakerrobot/easydetect-lab). 가중치 캐시는
 `~/.easydetect/`, 사내 미러는 `EASYDETECT_ASSETS_URL` 환경변수로 지정합니다.
 
 ## Credits

{easydetect-0.2.0 → easydetect-0.2.1}/README.md

@@ -103,18 +103,21 @@ faster on a small set, and `resume=True` picks a killed run back up.
 
 ## Label, train and watch it in a browser
 
-`platform/` is a web app on top of this package: drop images in, label them (the
-model drafts the boxes, you correct them), queue a training run, watch the curve,
-then run the result over a folder, a video or your webcam — on one machine, with
-nothing leaving it. A finished run hands you copy-ready code and a Hugging Face
-folder whose model card is written from the run.
+[easydetect lab](https://github.com/themakerrobot/easydetect-lab) is a web app on top of this package, in its own
+repository: drop images in, label them (the model drafts the boxes, you correct
+them), queue a training run, watch the curve, then run the result over a folder,
+a video or your webcam — on one machine, with nothing leaving it. A finished run
+hands you copy-ready code and a Hugging Face folder whose model card is written
+from the run.
 
 ```bash
-pip install -r platform/requirements.txt
-python platform/run.py          # http://<this machine>:8080
+git clone https://github.com/themakerrobot/easydetect-lab
+cd easydetect-lab
+pip install -r requirements.txt     # easydetect[train] from PyPI, and the web server
+python run.py                       # http://<this machine>:8080
 ```
 
-![the platform](https://raw.githubusercontent.com/themakerrobot/easydetect/main/docs/assets/platform.jpg)
+![easydetect lab](https://raw.githubusercontent.com/themakerrobot/easydetect-lab/main/docs/lab.jpg)
 
 ## Models
 
@@ -184,7 +187,7 @@ easydetect export  model=best.pt format=openvino half=true
 
 * [Using the model](https://github.com/themakerrobot/easydetect/blob/main/docs/usage.md) — sources, results, tracking, saving
 * [Training](https://github.com/themakerrobot/easydetect/blob/main/docs/training.md) — datasets, validation, export, CLI
-* [Platform](https://github.com/themakerrobot/easydetect/blob/main/platform/README.md) — the browser app: labelling, jobs, sharing
+* [easydetect lab](https://github.com/themakerrobot/easydetect-lab) — the browser app: labelling, jobs, sharing
 * [Weights](https://github.com/themakerrobot/easydetect/blob/main/docs/weights.md) — the mirror, building it, offline use
 * [Performance](https://github.com/themakerrobot/easydetect/blob/main/docs/performance.md) — measuring speed and improving it
 * [Design](https://github.com/themakerrobot/easydetect/blob/main/docs/design.md) — architecture and provenance
@@ -217,7 +220,7 @@ Runtime이 함께 들어가고 PyTorch는 없습니다. `pip install "easydetect
 `Detector("dfine-s", backend="onnxruntime")`로 ONNX Runtime(모든 CPU, 라즈베리파이
 포함)을 쓸 수 있습니다. 두 엔진의 결과는 같습니다.
 
-라벨링부터 학습·추론까지 브라우저로 하려면 `python platform/run.py`. 가중치 캐시는
+라벨링부터 학습·추론까지 브라우저로 하려면 [easydetect lab](https://github.com/themakerrobot/easydetect-lab). 가중치 캐시는
 `~/.easydetect/`, 사내 미러는 `EASYDETECT_ASSETS_URL` 환경변수로 지정합니다.
 
 ## Credits

{easydetect-0.2.0 → easydetect-0.2.1}/easydetect/__init__.py

@@ -24,7 +24,7 @@ needs numpy/opencv/openvino/pyyaml; training adds
 
 from __future__ import annotations
 
-__version__ = "0.2.0"
+__version__ = "0.2.1"
 
 from .errors import DownloadError, EasyDetectError, ModelNotFoundError
 from .metrics import BoxMetrics, DetMetrics

{easydetect-0.2.0 → easydetect-0.2.1}/easydetect/data/dataset.py

@@ -72,10 +72,10 @@ def _locate(root: Path, yaml_dir: Path, spec: str) -> Path:
     )
 
 
-def _list_images(root: Path, spec, yaml_dir: Path | None = None):
+def list_images(root: Path, spec, yaml_dir: Path | None = None):
     """Images for one split: a folder, a .txt list, or a list of either."""
     if isinstance(spec, (list, tuple)):
-        files = [f for s in spec for f in _list_images(root, s, yaml_dir)]
+        files = [f for s in spec for f in list_images(root, s, yaml_dir)]
         return list(dict.fromkeys(files))
     p = _locate(root, yaml_dir or root, str(spec))
     if p.is_dir():
@@ -109,7 +109,7 @@ class DetDataset(Dataset):
                 f"{data_yaml} has no '{split}:' entry. Add one — it may point at the "
                 f"training images, but then mAP only measures memorisation."
             )
-        self.files = _list_images(cfg["root"], cfg[split], cfg["yaml_dir"])
+        self.files = list_images(cfg["root"], cfg[split], cfg["yaml_dir"])
         if not self.files:
             raise FileNotFoundError(f"no images for split '{split}'")
 

{easydetect-0.2.0 → easydetect-0.2.1}/easydetect/data/labels.py

@@ -1,10 +1,10 @@
 # Apache-2.0
 """Where a label lives, and how one line of it reads — shared by everything.
 
-The trainer reads labels, the platform writes them, the exporter packs them.
+The trainer reads labels, easydetect lab writes them, the exporter packs them.
 If any two of those disagree on where ``a.jpg``'s boxes are, training runs on
 nothing and nobody is told. So there is one rule, here, and no torch import:
-the platform uses it at startup.
+the lab uses it at startup.
 """
 
 from __future__ import annotations

{easydetect-0.2.0 → easydetect-0.2.1}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "easydetect"
-version = "0.2.0"
+version = "0.2.1"
 description = "Easy real-time object detection you can ship: D-FINE with PyTorch training, OpenVINO inference on CPU/GPU/NPU, and a browser app to label and train. Apache-2.0 throughout."
 readme = "README.md"
 requires-python = ">=3.10"
@@ -84,5 +84,5 @@ ignore = ["B008", "E741"]
 "easydetect/utils/loss.py" = ["E501", "B905", "UP", "B006", "B007"]
 
 [tool.pytest.ini_options]
-testpaths = ["tests", "platform/tests"]
+testpaths = ["tests"]
 filterwarnings = ["ignore::DeprecationWarning"]

easydetect-0.2.1/tests/test_lab_api.py

@@ -0,0 +1,31 @@
+# Apache-2.0
+"""The names easydetect lab imports from this package.
+
+The lab lives in its own repository (themakerrobot/easydetect-lab) and runs on
+the PyPI release, so renaming or moving one of these breaks it only after a
+release. This test makes such a change visible here, before it ships.
+"""
+
+from __future__ import annotations
+
+import importlib
+
+import pytest
+
+LAB_IMPORTS = {
+    "easydetect": ["Detector", "__version__"],
+    "easydetect.data.dataset": ["DetDataset", "list_images", "load_data_yaml"],
+    "easydetect.data.labels": ["label_path", "label_row_to_box"],
+    "easydetect.downloads": ["MODEL_NAMES", "assets_url"],
+    "easydetect.plotting": ["draw_boxes"],
+    "easydetect.results": ["Boxes"],
+    "easydetect.sources": ["SourceLoader"],
+    "easydetect.validator": ["validate_torch"],
+}
+
+
+@pytest.mark.parametrize("module", sorted(LAB_IMPORTS))
+def test_the_lab_finds_what_it_imports(module):
+    mod = importlib.import_module(module)
+    missing = [name for name in LAB_IMPORTS[module] if not hasattr(mod, name)]
+    assert not missing, f"easydetect lab imports {missing} from {module}"

{easydetect-0.2.0 → easydetect-0.2.1}/tools/hub_README.md

@@ -105,7 +105,7 @@ model.export(format="openvino")                    # best.xml + best.bin + label
 
 `data.yaml` is the common images/ + labels/ layout — a Roboflow export works as
 it is. Or do it all in a browser: the
-[platform](https://github.com/themakerrobot/easydetect/tree/main/platform) collects
+[easydetect lab](https://github.com/themakerrobot/easydetect-lab) collects
 and labels images, trains, shows the numbers, and writes the upload folder for
 `models/` with its README filled in from the run.
 

{easydetect-0.2.0 → easydetect-0.2.1}/tools/make_toyset.py

@@ -7,7 +7,7 @@
 
 Every box is exact, shapes barely overlap, backgrounds vary, so a run that
 learns nothing here is broken rather than short of data. For trying the loop
-(train, val, export, predict, the platform), not for judging accuracy.
+(train, val, export, predict, the lab), not for judging accuracy.
 """
 import pathlib
 import random

easydetect-0.2.0/platform/README.md

@@ -1,199 +0,0 @@
-# easydetect platform
-
-Label images, train on them, watch the run, take the model away — in a browser,
-on one machine. It is a separate application that uses the `easydetect` package; the
-package itself stays a library with a CLI and no web dependencies.
-
-```bash
-pip install -r platform/requirements.txt   # in a clone, also: pip install -e ".[train]"
-python platform/run.py                 # http://<this machine>:8080, from any machine
-```
-
-![the platform](../docs/assets/platform.jpg)
-
-[PLAN.md](PLAN.md) is what this is meant to become, and what is missing today.
-
-The screens follow [themaker-ui](https://github.com/themakerrobot/themaker-ui),
-the design kit shared by themakerrobot's education tools: its stylesheet and the
-Pretendard font are copied unchanged into `static/css/` and `static/assets/`
-(CI checks the copy against the kit's v1.0.0 tag), and what is particular to
-this app sits in `static/css/app.css`. The **EN / 한** button at the top right
-switches between Korean and English; the choice is remembered in the browser,
-and a first visit follows the browser's language. Strings live in
-`static/js/i18n.js` — Korean is the source, with an English entry for each.
-Button names below are the English ones.
-
-## The loop
-
-1. **Collect** — four ways into a dataset, and all of them can add to one that
-   already exists: pick image files, upload a video and sample frames from it
-   (interval and cap are yours), upload a zip, or register a folder already on
-   this machine (`python platform/run.py label --source photos/ --names can,bottle`).
-   A YOLO-format export from a labelling service goes in as the zip it
-   downloaded as — Roboflow's *YOLOv8* / *YOLOv11* TXT, say — with its classes,
-   labels and train/valid split kept; *Statistics* and *Draw boxes* show it straight away.
-   Labels are stored where training reads them: in `labels/` beside an
-   `images/` folder, otherwise next to each picture, inside the folder.
-2. **Label** — *Draw boxes* on the dataset card. Drag to draw, drag inside a box to
-   move it, drag a corner to resize, <kbd>1</kbd>–<kbd>9</kbd> to set the class,
-   <kbd>Del</kbd> to remove, <kbd>Ctrl</kbd>+<kbd>Z</kbd> to undo. Saving is
-   automatic. The wheel zooms and the middle button (or <kbd>Space</kbd>) pans,
-   for boxes too small to place at fit-to-window. <kbd>C</kbd> copies the
-   previous image's boxes — a fixed camera repeats itself. Classes can be added
-   from the sidebar and show how many boxes each has, and the file list can hide
-   what is already labelled. A dataset spread over folders (`train/images`,
-   `valid/images`, …) can be browsed one folder at a time, and ←/→ stay in it;
-   background frames — a label with no boxes — are marked apart from pictures
-   that have boxes.
-3. **Keep the set honest** — *Statistics* counts boxes per class and flags what usually
-   bites: images with no label, labels with no boxes, specks under 0.1% of the
-   frame. *Duplicate* takes a snapshot before a risky relabel, *Merge selected* combines sets and
-   remaps class indices by name (index 0 rarely means the same thing in two
-   datasets), and an image can be dropped with its label from the labelling page.
-4. **Let the model do the first pass** — *Auto-label* fills the current image;
-   *Auto-label all* queues a job over every unlabelled image in the dataset.
-   Correcting boxes is far quicker than drawing them, and once you have a
-   `best.pt` you can point the auto-labeller at it.
-5. **Train** — a goal (빠르게 / 균형 / 정확하게) or your own model, epochs,
-   batch, backbone freezing, "stop when it stops improving" (patience) and
-   zoom/crop augmentation; image size and device under 고급 설정. A goal fills
-   only the fields you have not changed yourself. The header shows each NVIDIA GPU live —
-   utilisation, memory, temperature, power, read from `nvidia-smi` every two
-   seconds — and warns when the driver sees a GPU that PyTorch cannot use (a CPU
-   build of torch), since training would then run on the CPU. On a GPU the
-   presets batch four times larger. Jobs queue and run one
-   at a time; inside an epoch the job shows its batch, what is left and when it
-   is validating, and the loss and mAP curve updates per epoch; images are
-   decoded by loader processes (half the cores, at most eight —
-   `EASYDETECT_WORKERS` overrides); *Stop* stops within a batch;
-   *Train more* continues a finished run with more epochs in the same directory;
-   every run keeps a `train.log`, and a failure carries the end of it.
-   *Delete* removes a finished job with its weights, IR and log, and the runs
-   continued from it and the evaluations of it; it is refused while any of them
-   runs, or while a registered model uses those files. A run cut short — stopped,
-   crashed or interrupted by a restart — keeps its best epoch for evaluating,
-   downloading and registering.
-   *Run details* on the job shows what the run was set up with from its first
-   batch — starting weights, train/validation images and boxes per class,
-   trainable parameters, optimizer and schedule, augmentation, device, versions
-   — and, once it ends, the best epoch, time per epoch and final losses. The
-   trainer writes this to `run.json` in the run folder (`summary.json` repeats
-   it with the outcome), so it is there for scripts too.
-6. **See what it learned** — *Evaluate* scores the run on its validation split and
-   draws every one of those frames with the truth in green and the prediction in
-   its class colour, next to per-class AP. A single mAP says whether to keep
-   going; this says what to fix.
-7. **Run it on everything else** — *Predict* takes a dataset, a folder path on this
-   machine, or an uploaded video, and queues it like a training job. While it
-   works the job says how far it is ("이미지 54/300 · 상자 5개 · 남은 시간 약
-   43초") and shows frames as they are drawn; when done, the boxes per class and
-   the first 60 drawn frames are right there, with a zip of all of them (and
-   `results.json`) and, for a video, `annotated.mp4` to download. The model
-   picker lists your finished runs first — no need to register one to use it —
-   and auto-labelling drafts with the dataset's own latest run when it has one. *Webcam* opens the browser's camera and
-   posts a frame every 400 ms, so you see the model on live video even when the
-   server is somewhere else.
-8. **Take it away** — `best.pt`, the OpenVINO IR as a zip (exported when a run
-   finishes), `results.csv`. *Register model* gives the run a name (the IR if it has
-   one, otherwise the weights) and it then appears wherever a model is chosen —
-   auto-labelling, batch inference, the webcam. A `.pt`, `.onnx` or IR (its
-   `.xml` and `.bin` picked together) from elsewhere goes into the same list.
-   *Use it in code* on a finished run gives Python, command-line and "another
-   PC" snippets with this run's paths filled in, ready to copy.
-9. **Share it** — the *Hugging Face* tab builds the folder to upload: the IR,
-   `labels.txt`, optionally `best.pt`, and a `README.md` model card written
-   from the run itself — classes with their train/val box counts (and AP if
-   you evaluated), mAP, backbone, starting weights, the training settings and
-   hardware, and download-and-run code pointing at the repo and folder you
-   chose (by default the repo your weights come from, `models/<dataset>`).
-   It shows the `hf upload` command to run next. The plain IR zip carries the
-   same card.
-
-![labelling](../docs/assets/labeling.jpg)
-
-## Where things live
-
-Everything sits under `easydetect-platform/` in the directory you start from —
-`--data /some/path` or `$EASYDETECT_PLATFORM_HOME` moves it:
-
-```
-easydetect-platform/
-  platform.db          datasets, jobs, per-epoch numbers
-  datasets/<name>/     uploaded datasets (registered folders stay where they are)
-  runs/job<id>/        weights/, openvino/, results.csv, summary.json
-                       (inference jobs: images/, results.json, annotated.mp4)
-  models/<name>/       models uploaded from outside
-```
-
-## API
-
-The pages are only clients of these, so a script can do anything the UI does:
-
-| endpoint | |
-| --- | --- |
-| `POST /api/datasets` | multipart: `name`, `archive` (zip) |
-| `POST /api/datasets/images` | multipart: `name`, `files` (images), optional `dataset_id` |
-| `POST /api/datasets/video` | multipart: `name`, `video`, `every`, `max_frames` |
-| `GET /api/datasets/{id}/export` | the dataset as a zip |
-| `GET /api/datasets/{id}/stats` | boxes per class, and what looks wrong |
-| `DELETE /api/datasets/{id}/images/{index}` | drop one image and its label |
-| `POST /api/datasets/{id}/duplicate` | a copy to experiment on |
-| `POST /api/datasets/merge` | `{ids, name}` — one set, class indices remapped |
-| `DELETE /api/datasets/{id}` | |
-| `POST /api/datasets/local` | `{path, name, names}` — register a folder in place |
-| `GET /api/datasets` · `GET /api/datasets/{id}` | the second lists the images |
-| `GET`/`POST /api/datasets/{id}/labels/{index}` | read and write one image's boxes |
-| `POST /api/datasets/{id}/autolabel/{index}` | boxes for one image, from a model |
-| `POST /api/datasets/{id}/autolabel` | queue a pass over the whole dataset |
-| `POST /api/datasets/{id}/classes` | rename or add classes; data.yaml follows |
-| `GET /api/models` | the registry, plus the names that download themselves |
-| `POST /api/models` | `{job_id, name, note}` — register a finished run |
-| `POST /api/models/upload` | multipart: `name`, `files` (a .pt or .onnx, or an IR's .xml + .bin together), `classes` |
-| `DELETE /api/models/{id}` | forget it (files stay) |
-| `POST /api/predict` | multipart: `model`, `conf`, and one of `dataset_id`, `path`, `video` |
-| `POST /api/preview` | multipart: `model`, `conf`, `image` → annotated JPEG, one frame |
-| `POST /api/jobs` | `{dataset_id, model, epochs, imgsz, batch, freeze, device}` |
-| `GET /api/jobs` · `GET /api/jobs/{id}` | the second includes per-epoch rows |
-| `GET /api/gpu` | utilisation, memory, temperature and power per GPU, from nvidia-smi |
-| `GET /api/jobs/{id}/stream` | server-sent events: progress, then a status |
-| `POST /api/jobs/{id}/cancel` | |
-| `DELETE /api/jobs/{id}` | the job and its files, unless something still uses them |
-| `POST /api/jobs/{id}/resume` | `{add_epochs}` — continue a finished run |
-| `POST /api/jobs/{id}/evaluate` | `{conf}` — queue a scoring pass with pictures |
-| `GET /api/jobs/{id}/report` · `/eval/{name}` | the numbers, and the pictures |
-| `GET /api/jobs/{id}/download/{weights,openvino,results,log,predictions,video}` | |
-| `GET /api/jobs/{id}/download/huggingface` | `?repo=&folder=&pt=` — the upload folder, zipped, with its README |
-| `GET /api/jobs/{id}/modelcard` | the README.md that bundle would carry, and where it would go |
-| `POST /api/jobs/{id}/predict` | multipart: `image`, `conf` → annotated JPEG |
-
-## What it is not
-
-One process, one SQLite file, one folder. No accounts, no permissions, no
-quotas, no scheduler, and one job at a time. That is the point: a workstation or
-a mini PC beside a line needs no infrastructure, and the data never leaves it.
-
-When that stops being true — several people, several GPUs, work that must
-survive a restart — the things to change are the worker (onto a real queue) and
-the storage (onto a service), not the package underneath.
-
-Other limits worth knowing:
-
-* Boxes only. Polygons, keypoints, review workflows and team assignment are out
-  of scope; [CVAT](https://github.com/cvat-ai/cvat) (MIT) and
-  [Label Studio](https://github.com/HumanSignal/label-studio) (Apache-2.0) export
-  the same layout, so you can label there and train here.
-* Training runs in this process, so restarting the server ends a run. Jobs left
-  behind are marked failed on the next start rather than spinning forever.
-* It listens on every address (`0.0.0.0`) by default, so other machines can
-  open it — and it has no login: anyone who can reach the port can use it.
-  `--host 127.0.0.1` keeps it to this machine. Browsers only allow the webcam
-  on `https://` or `localhost`, so the same app is also served over HTTPS on
-  8443 (`--https-port`, 0 turns it off) with a certificate made on first start
-  and kept in `easydetect-platform/tls/`. The browser warns about it once; the
-  webcam tab opened over `http://` links to the HTTPS address.
-
-## Tests
-
-```bash
-pytest platform/tests
-```

easydetect-0.2.0/platform/tests/conftest.py

@@ -1,9 +0,0 @@
-# Apache-2.0
-"""The platform's modules live beside it, not on the install path."""
-
-from __future__ import annotations
-
-import sys
-from pathlib import Path
-
-sys.path.insert(0, str(Path(__file__).resolve().parents[1]))