attr-eomt 0.1.0__tar.gz

attr_eomt-0.1.0/.github/workflows/publish.yml

@@ -0,0 +1,41 @@
+name: publish
+
+# Builds the package and publishes to PyPI when a GitHub Release is published.
+# Uses PyPI Trusted Publishing (OIDC) — no API token stored as a secret.
+# One-time setup on PyPI: project Settings -> Publishing -> add a trusted publisher
+#   owner: imagra93   repo: attr-eomt   workflow: publish.yml   environment: pypi
+
+on:
+  release:
+    types: [published]
+  workflow_dispatch:        # allow manual runs (e.g. re-publishing an existing release)
+
+jobs:
+  build:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v5
+      - name: Install uv
+        uses: astral-sh/setup-uv@v6
+      - name: Build sdist + wheel
+        run: uv build
+      - name: Check metadata
+        run: uvx twine check dist/*
+      - uses: actions/upload-artifact@v4
+        with:
+          name: dist
+          path: dist/
+
+  publish:
+    needs: build
+    runs-on: ubuntu-latest
+    environment: pypi
+    permissions:
+      id-token: write  # required for Trusted Publishing
+    steps:
+      - uses: actions/download-artifact@v4
+        with:
+          name: dist
+          path: dist/
+      - name: Publish to PyPI
+        uses: pypa/gh-action-pypi-publish@release/v1

attr_eomt-0.1.0/.github/workflows/tests.yml

@@ -0,0 +1,32 @@
+name: tests
+
+on:
+  push:
+    branches: [main]
+  pull_request:
+
+jobs:
+  test:
+    runs-on: ubuntu-latest
+    strategy:
+      fail-fast: false
+      matrix:
+        python-version: ["3.10", "3.11", "3.12"]
+    steps:
+      - uses: actions/checkout@v5
+
+      - name: Install uv
+        uses: astral-sh/setup-uv@v6
+        with:
+          enable-cache: true
+
+      - name: Create venv on Python ${{ matrix.python-version }}
+        run: uv venv --python ${{ matrix.python-version }}
+
+      - name: Install project (CPU torch)
+        env:
+          UV_TORCH_BACKEND: cpu
+        run: uv pip install -e ".[dev]"
+
+      - name: Run tests
+        run: uv run --no-sync pytest -q

attr_eomt-0.1.0/.gitignore

@@ -0,0 +1,31 @@
+# Python
+__pycache__/
+*.py[cod]
+*.egg-info/
+.eggs/
+build/
+dist/
+.pytest_cache/
+.ipynb_checkpoints/
+
+# Virtual envs
+.venv/
+venv/
+env/
+
+# Training artifacts
+runs/
+weights/
+data/
+*.pt
+*.pth
+
+# Datasets (auto-downloaded)
+datasets/
+data/coco/
+scripts/sample_images/
+
+# Editor / OS
+.idea/
+.vscode/
+.DS_Store

attr_eomt-0.1.0/LICENSE

@@ -0,0 +1,201 @@
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or Derivative
+          Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and do
+          not modify the License. You may add Your own attribution notices
+          within Derivative Works that You distribute, alongside or as an
+          addendum to the NOTICE text from the Work, provided that such
+          additional attribution notices cannot be construed as modifying
+          the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "[]"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright [yyyy] [name of copyright owner]
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.

attr_eomt-0.1.0/PKG-INFO

@@ -0,0 +1,307 @@
+Metadata-Version: 2.4
+Name: attr-eomt
+Version: 0.1.0
+Summary: Standalone EoMT (Encoder-only Mask Transformer) for instance segmentation, with DINOv2 init, COCO training/validation and inference.
+Project-URL: Homepage, https://github.com/imagra93/attr-eomt
+Project-URL: Repository, https://github.com/imagra93/attr-eomt
+Project-URL: Issues, https://github.com/imagra93/attr-eomt/issues
+Author: attr-eomt contributors
+License-Expression: Apache-2.0
+License-File: LICENSE
+Keywords: coco,dinov2,eomt,instance-segmentation,transformers
+Classifier: License :: OSI Approved :: Apache Software License
+Classifier: Programming Language :: Python :: 3
+Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
+Requires-Python: >=3.10
+Requires-Dist: huggingface-hub>=0.23.0
+Requires-Dist: matplotlib>=3.5.0
+Requires-Dist: numpy>=1.19.0
+Requires-Dist: pillow>=9.1.0
+Requires-Dist: pycocotools>=2.0.0
+Requires-Dist: pyyaml>=6.0
+Requires-Dist: requests>=2.25.0
+Requires-Dist: scipy>=1.7.0
+Requires-Dist: torch>=2.4.0
+Requires-Dist: torchvision>=0.19.0
+Requires-Dist: tqdm>=4.65.0
+Requires-Dist: transformers>=5.1.0
+Provides-Extra: dev
+Requires-Dist: build>=1.0; extra == 'dev'
+Requires-Dist: pytest>=7.0; extra == 'dev'
+Requires-Dist: twine>=5.0; extra == 'dev'
+Provides-Extra: logging
+Requires-Dist: tensorboard>=2.10; extra == 'logging'
+Requires-Dist: wandb>=0.15; extra == 'logging'
+Description-Content-Type: text/markdown
+
+# attr-eomt
+
+Standalone **EoMT** (Encoder-only Mask Transformer) for **instance segmentation**,
+with one feature that sets it apart: **secondary per-instance classification heads**
+("auxiliary classes"). Alongside the usual mask + class output, the model predicts
+**one or several independent attributes for every detected instance** — and they
+train and infer for free on top of segmentation, without inflating the primary
+class space.
+
+The name reflects exactly that: **attr-eomt** is EoMT extended with per-instance
+**attr**ibute heads.
+
+EoMT itself is a DINOv2-with-registers ViT whose last few transformer blocks are
+augmented with learnable queries producing mask-classification output
+(Mask2Former-style). This package builds it in three sizes, initializes the encoder
+from DINOv2, and provides training, per-epoch COCO-mAP validation, and
+inference/rendering — all behind a small `EoMT` class. It is a clean-room,
+Apache-2.0-compatible reimplementation; weights you train are yours to release.
+
+```python
+from eomt import EoMT
+
+model = EoMT("l")                            # fresh large model (DINOv2 backbone)
+model.train(data="coco", epochs=50)          # COCO 2017 auto-downloads if missing
+
+model = EoMT("runs/train/eomt-l")            # reload a run — size/classes/heads auto-detected
+model.predict("images/", plot=True)          # render masks + per-instance attributes
+```
+
+---
+
+## ⭐ Auxiliary classes (secondary per-instance classification)
+
+The primary task (instance segmentation over `nc` classes) is **unchanged**. On top
+of it you can attach **one or several independent secondary classifiers** — a
+per-instance *attribute* predicted for each detected mask, read straight from that
+query's embedding. You can have as many heads as your data defines.
+
+This answers the "classes **and** subclasses per instance" need by encoding a
+*subclass per instance* rather than flattening to a `class × subclass` product space
+(which would wreck Hungarian matching and thin out per-class statistics). Because
+EoMT is **NMS-free**, two overlapping same-class instances stay two distinct queries;
+the attribute head separates them from their embeddings.
+
+### Example: a bowl of fruit
+
+One model segments each fruit (primary classes `apple` / `banana` / `orange` /
+`pear`) and, for **every** detection, reads off two **independent** attribute heads —
+`ripeness` (`unripe` / `turning` / `ripe`) and a quality `grade` (`A` / `B`). The
+renderer prints the primary class + score on the first row and each attribute + its
+confidence on the row beneath it.
+
+![A bowl of fruit; each box labelled with its fruit class plus a ripeness and a grade](docs/examples/example_1.png)
+
+`ripeness` and `grade` are *orthogonal* — they vary independently — which is exactly
+the case that's awkward to fold into the primary class space. And because EoMT is
+NMS-free, two apples (one ripe, one unripe) and two oranges stay four distinct
+queries, each with its own `ripeness` and `grade`.
+
+> The image above is a **simulation**: a stock photo with hand-placed detections fed
+> through the package's own renderer ([`eomt.visualize.draw_instances`](eomt/visualize.py))
+> to show the output format — not a trained model's predictions. The same pattern
+> fits any "class **plus** per-instance sub-labels" task: **retail shelves → product +
+> facing**, **cells / leaves → type + health**, **apparel → garment + pattern**.
+
+### How it works
+
+- **Embedding source.** Each head reads the per-query embedding — the input to EoMT's
+  `class_predictor`, captured with a forward hook (`[B, Q, hidden]`).
+- **Matching.** Supervision reuses EoMT's *own* Hungarian matcher
+  (`model.eomt.criterion.matcher`), so every attribute is trained on the **same**
+  query→GT assignment the detection loss used. The attribute is read *after* matching.
+- **Loss.** Cross-entropy per head over matched queries, summed across heads and
+  scaled by `aux_w` (default `1.0`), added to the segmentation loss. Empty-match
+  batches contribute a graph-preserving zero.
+- **Checkpoint selection stays `segm/mAP`.** The attribute "rides along": its per-head
+  matched-query train accuracy is shown live and written to `metrics.csv`, but never
+  drives `best.pt`.
+- **Inference.** Each result attaches `aux = {head: {"ids", "probs"}}` for the kept
+  detections, and `predict(plot=True)` renders each attribute next to the class label
+  using names stored in the checkpoint.
+
+### Data format (auto-discovered from the COCO JSON)
+
+Attributes live **inside the COCO annotations** — each annotation is already a
+per-instance object, so alignment is automatic and `pycocotools` still parses it. Two
+additions to a standard COCO file:
+
+```jsonc
+{
+  "categories": [ {"id": 1, "name": "apple"}, {"id": 2, "name": "banana"} ],
+
+  "attributes": [                                       // NEW, top-level: per-head vocab(s)
+    {"name": "ripeness", "categories": [{"id": 0, "name": "unripe"},
+                                        {"id": 1, "name": "turning"},
+                                        {"id": 2, "name": "ripe"}]},
+    {"name": "grade",    "categories": [{"id": 10, "name": "A"},
+                                        {"id": 20, "name": "B"}]}
+  ],
+
+  "annotations": [
+    { "id": 1, "image_id": 42, "category_id": 1,
+      "segmentation": [...], "bbox": [...], "area": 1234, "iscrowd": 0,
+      "attributes": {"ripeness": 1, "grade": 20} }       // NEW, per instance: {head: raw_id}
+  ]
+}
+```
+
+- The top-level `attributes` list defines each head's vocabulary; raw ids are remapped
+  to a contiguous `0..n-1` per head (so the non-contiguous `grade` ids `10`/`20` become
+  `0`/`1`). `categories` may be omitted, in which case the id set is inferred.
+- Per-annotation `attributes` is a `{head: raw_id}` map; a missing value defaults to `0`.
+- A JSON with **no** `attributes` ⇒ detection-only, behaving exactly as before.
+
+No YAML changes are needed — heads (count, classes, names) are discovered straight from
+the JSON, the same as `nc`. A tiny, self-contained example (two heads, including a
+non-contiguous id set) lives in [sample_data/](sample_data/).
+
+---
+
+## Sizes
+
+| size | backbone        | hidden | layers | heads | queries |
+|------|-----------------|--------|--------|-------|---------|
+| `s`  | DINOv2-small    | 384    | 12     | 6     | 100     |
+| `b`  | DINOv2-base     | 768    | 12     | 12    | 200     |
+| `l`  | DINOv2-large    | 1024   | 24     | 16    | 200     |
+
+Default input is a patch-14-aligned square (`644 = 14 × 46`) so DINOv2 weights load 1:1.
+
+## Install
+
+```bash
+pip install attr-eomt                  # from PyPI
+pip install "attr-eomt[logging]"       # + tensorboard/wandb
+```
+
+Or from source (editable, for development):
+
+```bash
+pip install -e ".[dev]"                # [dev] adds pytest/build/twine
+```
+
+## Pretrained weights
+
+Checkpoints are hosted on the **Hugging Face Hub**, not bundled in the wheel.
+Load one in a single line — it downloads once and is cached for later runs:
+
+```python
+from eomt import EoMT
+
+model = EoMT.from_pretrained("imagra93/eomt-l-coco")   # downloads + caches
+results = model.predict("images/", plot=True)
+```
+
+Equivalently, an `hf://` reference works anywhere a checkpoint path is accepted:
+
+```python
+EoMT("hf://imagra93/eomt-l-coco/model.pt").val(data="coco")
+```
+
+`from_pretrained` accepts `filename=` (which checkpoint in the repo, default
+`model.pt`), `revision=` (branch/tag/commit) and `device=`.
+
+To publish your own trained weights to the Hub (creates the repo if needed; needs
+`huggingface-cli login` or `HF_TOKEN`):
+
+```python
+EoMT("runs/train/eomt-l").push_to_hub("your-username/eomt-l-coco")   # private by default
+```
+
+## Quickstart
+
+Everything goes through one class. Initialize it from a **size** (a fresh model with a
+pretrained DINOv2 backbone) or from a **checkpoint / run folder** (size, classes, image
+size and any auxiliary heads are auto-detected):
+
+```python
+from eomt import EoMT
+
+# Train on COCO 2017 (auto-downloaded on first run). batch 4 with the default
+# nominal_batch 16 accumulates to an effective batch of 16.
+EoMT("l").train(data="coco", epochs=50, batch=4)
+
+# ...or any COCO-format dataset (point at its data.yaml):
+EoMT("s").train(data="sample_data/data.yaml", epochs=1, batch=1)
+
+# Validate a checkpoint (COCO segm + bbox mAP):
+metrics = EoMT("runs/train/eomt-l").val(data="coco")
+
+# Predict + render on an image or a folder; results carry boxes/scores/classes/masks
+# (+ `aux` for models with auxiliary heads). plot=True writes annotated images:
+results = EoMT("runs/train/eomt-l").predict("images/", plot=True)
+```
+
+Ready-to-run wrappers live in [scripts/](scripts/):
+
+```bash
+python scripts/train.py --size l --epochs 50 --batch 4     # COCO by default
+python scripts/val.py     runs/train/eomt-l
+python scripts/predict.py runs/train/eomt-l                 # uses scripts/sample_images
+```
+
+To resume or fine-tune, initialize from the checkpoint:
+
+```python
+EoMT("runs/train/eomt-l").train(data="coco", resume=True)   # continue the run
+EoMT("runs/train/eomt-l").train(data="my.yaml", epochs=20)  # warm-start (fine-tune)
+```
+
+### Useful `train()` options
+
+Passed as keyword arguments to `model.train(...)`:
+
+| arg | default | effect |
+|------|---------|--------|
+| `nominal_batch` / `accum` | `16` / `0` | gradient accumulation to an **effective batch** of `nominal_batch` (EoMT recipe is 16); `accum=N` sets the step count explicitly |
+| `ema` / `ema_decay` / `ema_tau` | `True` / `0.9999` / `2000` | validate & export `best.pt` from an **EMA** of the weights; decay + its warmup ramp |
+| `llrd` | `0.85` | **layer-wise LR decay** on the DINOv2 backbone (`1.0` = flat `backbone_lr_mult`) |
+| `min_scale` / `max_scale` | `0.1` / `2.0` | **Large-Scale Jitter** range (legacy stretch-style: `0.5` / `1.0`) |
+| `letterbox` | `True` | aspect-preserving **letterbox** eval (vs legacy square stretch); recorded in the checkpoint |
+| `flip_prob` | `0.5` | horizontal-flip probability. **Set `0`** for datasets with a left/right attribute — hflip mirrors pixels without swapping the laterality label |
+| `mask_anneal` / `mask_anneal_start` / `mask_anneal_end` | `True` / `0.0` / `0.9` | anneal masked attention `1→0` over this fraction of training (EoMT recipe) |
+| `aux_w` | `1.0` | weight on the summed secondary-head loss |
+
+## Training recipe
+
+Defaults follow the EoMT/Mask2Former fine-tuning recipe; each piece is a keyword
+argument, so the legacy behaviour is one override away.
+
+- **Effective batch via gradient accumulation.** LR / weight-decay / clip are tuned for
+  an effective batch of 16, so training accumulates `round(nominal_batch / batch)`
+  micro-batches per optimizer step. EoMT is a ViT (LayerNorm, no BatchNorm), so this is
+  ~equivalent to a true large batch at a fraction of the memory.
+- **EMA weights.** A moving average is validated and saved as `best.pt`; `last.pt` holds
+  the live weights plus optimizer and EMA state for exact resume.
+- **Large-Scale Jitter (LSJ).** Training resizes aspect-preserving over
+  `[min_scale, max_scale]` then crops/pads to the square input — a strong scale aug.
+- **Letterbox eval.** Validation/inference resize the long side and pad to a square; the
+  padding is cropped back out in postprocessing. The mode is stored per-checkpoint so
+  `val`/`predict` match training automatically.
+- **Optimizer.** AdamW with no weight decay on norms/biases/embeddings and layer-wise LR
+  decay on the backbone (`llrd`, deeper layers get a higher LR).
+- **Tunable objective.** The matcher/loss weights (`class_weight`, `mask_weight`,
+  `dice_weight`, `no_object_weight`), PointRend sampling (`train_num_points`) and
+  mask-head depth (`num_upscale_blocks`) are arguments, persisted in the checkpoint so a
+  tuned objective rebuilds on reload.
+- **Masked-attention annealing.** The masked-attention probability is annealed `1 → 0`
+  over `[mask_anneal_start, mask_anneal_end]` of training, so the final stretch trains
+  **mask-free** and matches efficient (mask-less) inference. Validation and
+  checkpointing run mask-free (deterministic).
+
+## What's included
+
+- Architecture (`s`/`b`/`l`) + DINOv2 init — `eomt.model`, `eomt.config`
+- **Secondary per-instance attribute heads** — `eomt.aux_cls`, `eomt.config.AuxHeadSpec`
+- COCO-format datasets (incl. per-instance attributes), Large-Scale Jitter + letterbox
+  augmentations (torchvision v2), autodownload — `eomt.data`
+- Training loop (AdamW + layer-wise LR decay + no-WD groups, cosine warmup, gradient
+  accumulation, AMP, EMA weights, masked-attention annealing, aux-head loss, resume,
+  per-epoch `metrics.csv`) — `eomt.engine.train`, `eomt.ema`
+- COCO-mAP validation (`pycocotools`) with letterbox-aware mask remapping — `eomt.engine.validate`
+- Mask2Former-style scoring (class confidence × mask objectness) — `eomt.postprocess`
+- Inference + rendering (with attribute labels) — `eomt.engine.predict`, `eomt.visualize`
+
+## Not included
+
+Deployment-format export (ONNX / TensorRT / etc.) is intentionally out of scope. A
+detect (true box head) family and a semantic-segmentation family are planned; the code
+carries a `family` parameter so they can be added without API churn, but only
+`instance` is implemented today.