Model card: cut what the code repo already documents

Removes the sections duplicated from the GitHub README (usage command block, class lists, beta-convention explanation, licence detail, full citation BibTeX) and links there instead. Keeps only what is specific to these artifacts: which seed and epoch, checkpoint dict layout, the non-portable TensorRT engine, and the model-card limitations. 198 -> 124 lines.

README.md

@@ -21,79 +21,47 @@ metrics:
 Released weights for **AffKernel**: single-pass, NMS-free affordance
 segmentation. RT-DETR object queries generate per-query dynamic convolution
 kernels that decode per-object affordance masks from one shared high-resolution
-map. A detector tells a robot *what* an object is; AffKernel also tells it
-*where* to act on it.
+map.
 
-**Documentation lives in the code repository. This card covers the artifacts
-only.**
-
-| | |
-|---|---|
-| Code, results, deployment guide | https://github.com/anh0001/affkernel |
-| Reproduction guide | [`docs/reproduction.md`](https://github.com/anh0001/affkernel/blob/main/docs/reproduction.md) |
-| Dataset setup | [`docs/datasets.md`](https://github.com/anh0001/affkernel/blob/main/docs/datasets.md) |
+> **This card describes the artifacts only.** Method, results, ablations,
+> deployment benchmarks, protocol caveats, licensing detail and citation all
+> live in the code repository and are deliberately not repeated here:
+> **https://github.com/anh0001/affkernel**
+> ([reproduction guide](https://github.com/anh0001/affkernel/blob/main/docs/reproduction.md),
+> [dataset setup](https://github.com/anh0001/affkernel/blob/main/docs/datasets.md))
 
 ## What this checkpoint is
 
-Trained on IIT-AFF (6,184 train images), 72 epochs, seed 42, last-epoch EMA
-weights, fp32.
+IIT-AFF, 72 epochs, **seed 42**, last-epoch EMA weights, fp32.
 
-| Metric (IIT-AFF test, 2,651 images) | Value |
+| | |
 |---|---:|
-| `F_beta^w` (beta^2 = 1), this checkpoint | **0.8685** |
-| `F_beta^w` (beta^2 = 0.3), this checkpoint | 0.8582 |
-| Model-only latency, 640x640, fp32, RTX 6000 Ada | 16.3 ms |
-
-For context the three-seed mean is 0.8675 ± 0.0009; this is seed 42, the paper's
-anchor seed and the highest of the three (seed 7: 0.8668, seed 123: 0.8673). The
-epoch was fixed a priori (last epoch, EMA) rather than selected on the test set.
+| `F_beta^w` (beta^2 = 1) | **0.8685** |
+| `F_beta^w` (beta^2 = 0.3) | 0.8582 |
 
-Two `beta` conventions circulate in this literature, so both are given. Never
-compare a `beta^2 = 1` number against a `beta^2 = 0.3` one.
+This is one seed, not the headline mean: the three-seed figure is
+0.8675 ± 0.0009, and seed 42 is the highest of the three (7: 0.8668,
+123: 0.8673). Both beta conventions are given because a `beta^2 = 1` number
+must never be compared against a `beta^2 = 0.3` one. Read the accuracy as
+exploratory rather than confirmatory: IIT-AFF ships no validation split, so the
+same test set also informed the recipe. Full reasoning in the repository README.
 
-> **These results are exploratory, not confirmatory.** IIT-AFF ships no
-> validation split, so the same test set that produced this number also informed
-> the recipe as it evolved and the choice of reported configuration. Fixing the
-> last epoch a priori prevents peak-checkpoint selection but not that broader
-> adaptive use. Read it as evidence from a single benchmark, not as an estimate
-> of generalisation.
-
-Baseline comparisons, the readout-resolution ladder, the full deployment
-benchmarks and the ablations are all in the
-[repository README](https://github.com/anh0001/affkernel#readme). Peer latencies
-there are measured on each method's own hardware and are not normalised, so no
-speed ratio between them is meaningful.
-
-## Usage
+## Load it
 
 ```bash
-git clone https://github.com/anh0001/affkernel.git
-cd affkernel
-pip install -r requirements.txt huggingface_hub
-
 hf download anhrisn/affkernel-iit-aff \
   affkernel_iit_r50vd_stride2_deepsup_seed42.pth --local-dir weights/
-
-python tools/infer.py \
-  -c configs/rtdetr/rtdetr_r50vd_6x_iit_v3_stride2_deepsup.yml \
-  -r weights/affkernel_iit_r50vd_stride2_deepsup_seed42.pth \
-  --input path/to/image.jpg \
-  --output outputs/
 ```
 
-`--input` takes an image or a directory; `--output` is a directory, written as
-`<stem>_aff.png`. Add `--device cpu` to run without a GPU.
-
-The first run downloads the ImageNet-pretrained ResNet-50vd backbone from the
-RT-DETR release artefacts, so it needs network access. Set
-`PResNet.pretrained: False` for an offline deployment.
+Inference, evaluation and deployment commands are in the
+[repository README](https://github.com/anh0001/affkernel#readme).
 
 ## Files
 
 | File | What it is |
 |---|---|
 | `affkernel_iit_r50vd_stride2_deepsup_seed42.pth` | The model. fp32 EMA weights, 174 MB. |
-| `backbone_fp16.plan` | Optional. TensorRT fp16 **backbone-only** engine, 46.4 MiB. Not portable, not usable on its own. |
+| `backbone_fp16.plan` | Optional TensorRT fp16 **backbone-only** engine, 46.4 MiB. Not portable, not usable on its own. |
 
 `deepsup` in the filename is the historical identifier for the auxiliary readout
 losses used during training; it is kept so the documented commands keep working.
@@ -111,6 +79,10 @@ are not included. **Do not repack this into a bare `{"model": ...}` dict**: unde
 an EMA-enabled config the solver would then evaluate a freshly initialised EMA
 module and score near zero.
 
+The first run fetches the ImageNet-pretrained ResNet-50vd backbone from the
+RT-DETR release artefacts, so it needs network access. Set
+`PResNet.pretrained: False` for an offline deployment.
+
 ### `backbone_fp16.plan`
 
 The dynamic-kernel affordance head cannot be exported to ONNX, but the backbone
@@ -128,71 +100,25 @@ CUDA-graphed against the engine's output buffers. Built on Jetson AGX Orin
 > minutes, and it is the supported path. See
 > [Deployment](https://github.com/anh0001/affkernel#deployment-on-nvidia-jetson).
 
-## Classes
-
-**Objects (10):** bowl, tvm, pan, hammer, knife, cup, drill, racket, spatula, bottle
-
-**Affordances (9):** contain, cut, display, engine, grasp, hit, pound, support, w-grasp
-
 ## Intended use and limitations
 
-Intended for research on affordance perception and perception-guided grasping.
+Research on affordance perception and perception-guided grasping.
 
-- Closed vocabulary of 10 object and 9 affordance classes; it will not
-  generalise to unseen categories.
+- Closed vocabulary: 10 object and 9 affordance classes, listed in
+  [`docs/datasets.md`](https://github.com/anh0001/affkernel/blob/main/docs/datasets.md).
+  It will not generalise to unseen categories.
 - IIT-AFF is a tabletop dataset. Other viewpoints, lighting or clutter regimes
   are untested.
 - The residual error is dominated by **missed detections** rather than by mask
-  quality. On instances the detector does fire on, mask quality reaches 0.893;
-  that figure excludes detection misses, so it is not protocol-matched to
-  published full-set numbers and is not a state-of-the-art claim.
-- The headline latency is fp32 on an RTX 6000 Ada. fp16, CUDA graphs and the
-  TensorRT backbone are characterised on that machine and on a Jetson AGX Orin
-  (23.1 FPS end to end within 0.23 GiB) and cost at most 0.0007 `F_beta^w`,
-  below the seed-to-seed spread.
+  quality.
 - Not validated for safety-critical deployment. A predicted grasp region is a
   perception cue, not a guarantee of a safe grasp.
 
-## License and attribution
-
-These weights are Apache-2.0, matching the RT-DETR components and the
-ImageNet-pretrained backbone they were initialised from. The repository's own
-source contributions are MIT; see
-[`THIRD_PARTY_LICENSES.md`](https://github.com/anh0001/affkernel/blob/main/THIRD_PARTY_LICENSES.md).
-
-The **IIT-AFF dataset is not redistributed** here or in the code repository. It
-states no licence and its authors request citation of the original paper. Obtain
-it from https://sites.google.com/site/iitaffdataset/.
-
-## Citation
-
-```bibtex
-@article{risnumawan2026affkernel,
-  title   = {AffKernel: High-Resolution Readout for Real-Time Visual
-             Affordance Segmentation},
-  author  = {Risnumawan, Anhar and Aji, Achmad Fahrul and
-             Fatahillah, Teuku Zikri and Kubota, Naoyuki},
-  journal = {Expert Systems with Applications},
-  year    = {2026},
-  note    = {Under review}
-}
-
-@inproceedings{nguyen2017object,
-  title     = {Object-Based Affordances Detection with Convolutional Neural
-               Networks and Dense Conditional Random Fields},
-  author    = {Nguyen, Anh and Kanoulas, Dimitrios and Caldwell, Darwin G. and
-               Tsagarakis, Nikos G.},
-  booktitle = {IEEE/RSJ International Conference on Intelligent Robots and
-               Systems (IROS)},
-  year      = {2017}
-}
-
-@inproceedings{lv2024detrs,
-  title     = {DETRs Beat YOLOs on Real-time Object Detection},
-  author    = {Lv, Wenyu and Zhao, Yian and Xu, Shangliang and Wei, Jinman and
-               Wang, Guanzhong and Cui, Cheng and Du, Yuning and Dang, Qingqing
-               and Liu, Yi},
-  booktitle = {CVPR},
-  year      = {2024}
-}
-```
+## License
+
+Weights are Apache-2.0, matching the RT-DETR components and the
+ImageNet-pretrained backbone they were initialised from. The **IIT-AFF dataset
+is not redistributed** here; obtain it from
+https://sites.google.com/site/iitaffdataset/. Full attribution and the citation
+BibTeX are in the
+[repository](https://github.com/anh0001/affkernel/blob/main/THIRD_PARTY_LICENSES.md).