Update AnimeTimm organization card

Add a DeepGHS-stewarded organization profile with dbv4-full dataset details, model tables, reproducible chart assets, and local maintenance guidance.

.gitattributes

@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+assets/dbv4-full-model-snapshot.png filter=lfs diff=lfs merge=lfs -text

AGENTS.md

@@ -0,0 +1,45 @@
+# AnimeTimm Organization Card Maintenance
+
+This repository backs the Hugging Face organization card for `animetimm`.
+
+## Scope
+
+- Keep this repository focused on the public organization profile shown at `https://huggingface.co/animetimm`.
+- Maintain `README.md`, public visual assets under `assets/`, checked-in source data under `data/`, and reproducible helper scripts under `scripts/`.
+- Do not add private credentials, local absolute paths, unpublished benchmark data, or private repository details.
+
+## Stewardship Language
+
+- Describe AnimeTimm as a DeepGHS project.
+- Link to `https://huggingface.co/deepghs`, `https://github.com/deepghs`, and `https://github.com/deepghs/animetimm` when discussing ownership or upstream engineering work.
+- Keep the tone lightly academic and anime-hobbyist friendly: technically clear, but not corporate or overly formal.
+
+## Model Zoo Policy
+
+- The organization card should currently feature only the main `dbv4-full` model line.
+- Do not list `dbv4a`, `witha`, `e621`, `pexels`, private, or experimental branches on the homepage unless the project direction changes explicitly.
+- Keep two model tables: top 5 by `Macro@Best F1`, and one best representative per backbone family.
+- Store model metrics in `data/dbv4_full_models.csv`; generated JSON files and charts should be derived from that source.
+
+## Regeneration
+
+Run the following command from the repository root after updating `data/dbv4_full_models.csv`:
+
+```bash
+python scripts/build_org_card.py
+```
+
+Review the generated files before committing:
+
+- `assets/animetimm-banner.png`
+- `assets/dbv4-full-model-snapshot.png`
+- `data/dbv4_full_models.json`
+- `data/featured_models.json`
+
+## Commit Policy
+
+Use the repository owner's Hugging Face account/token when pushing updates. Commit messages should be concise English summaries, for example:
+
+```text
+Update AnimeTimm organization card
+```

CLAUDE.md

@@ -0,0 +1,45 @@
+# AnimeTimm Organization Card Maintenance
+
+This repository backs the Hugging Face organization card for `animetimm`.
+
+## Scope
+
+- Keep this repository focused on the public organization profile shown at `https://huggingface.co/animetimm`.
+- Maintain `README.md`, public visual assets under `assets/`, checked-in source data under `data/`, and reproducible helper scripts under `scripts/`.
+- Do not add private credentials, local absolute paths, unpublished benchmark data, or private repository details.
+
+## Stewardship Language
+
+- Describe AnimeTimm as a DeepGHS project.
+- Link to `https://huggingface.co/deepghs`, `https://github.com/deepghs`, and `https://github.com/deepghs/animetimm` when discussing ownership or upstream engineering work.
+- Keep the tone lightly academic and anime-hobbyist friendly: technically clear, but not corporate or overly formal.
+
+## Model Zoo Policy
+
+- The organization card should currently feature only the main `dbv4-full` model line.
+- Do not list `dbv4a`, `witha`, `e621`, `pexels`, private, or experimental branches on the homepage unless the project direction changes explicitly.
+- Keep two model tables: top 5 by `Macro@Best F1`, and one best representative per backbone family.
+- Store model metrics in `data/dbv4_full_models.csv`; generated JSON files and charts should be derived from that source.
+
+## Regeneration
+
+Run the following command from the repository root after updating `data/dbv4_full_models.csv`:
+
+```bash
+python scripts/build_org_card.py
+```
+
+Review the generated files before committing:
+
+- `assets/animetimm-banner.png`
+- `assets/dbv4-full-model-snapshot.png`
+- `data/dbv4_full_models.json`
+- `data/featured_models.json`
+
+## Commit Policy
+
+Use the repository owner's Hugging Face account/token when pushing updates. Commit messages should be concise English summaries, for example:
+
+```text
+Update AnimeTimm organization card
+```

README.md

@@ -1,10 +1,94 @@
 ---
 title: README
-emoji: 👀
-colorFrom: gray
-colorTo: green
+emoji: 🎨
+colorFrom: pink
+colorTo: indigo
 sdk: static
 pinned: false
 ---
 
-Edit this `README.md` markdown file to author your organization card.
+<img src="./assets/animetimm-banner.png" alt="AnimeTimm banner" width="100%">
+
+# AnimeTimm
+
+**AnimeTimm is a DeepGHS project for training, testing, and sharing `timm`-based vision models for anime-style and illustration-focused image tagging.**
+
+It is part research playground, part anime-fan workshop: we care about reproducible datasets, model cards, ONNX exports, and practical demos, but the models are also built for people who actually work with 2D art, tags, characters, and visual search.
+
+## Project Stewardship
+
+AnimeTimm is produced and maintained by the [DeepGHS](https://huggingface.co/deepghs) team and contributors.
+
+This Hugging Face organization is the focused publishing home for AnimeTimm releases: model checkpoints, selected training datasets, and interactive Spaces. The upstream engineering work is connected to the [DeepGHS GitHub organization](https://github.com/deepghs), including the [`deepghs/animetimm`](https://github.com/deepghs/animetimm) repository.
+
+## What We Build
+
+- `timm`-based image tagging and classification models for anime-style images.
+- Training datasets prepared for large-scale tagger experiments.
+- PyTorch, Safetensors, and ONNX artifacts where available.
+- Playgrounds and ranklists for trying models and comparing outputs.
+
+## Featured Dataset
+
+### [`danbooru-wdtagger-v4-w640-ws-full`](https://huggingface.co/datasets/animetimm/danbooru-wdtagger-v4-w640-ws-full)
+
+The main public dataset release used by the `dbv4-full` model family. It is a Danbooru-derived WebDataset build for large-scale anime-style multi-label tagging, with images resized so `min(width, height) <= 640`.
+
+| Split | Images | Total Size |
+| --- | ---: | ---: |
+| train | 5,321,713 | 318 GB |
+| test | 295,926 | 17.7 GB |
+| val | 296,957 | 17.8 GB |
+| **total** | **5,914,596** | **353.5 GB** |
+
+Each sample contains the image as `webp` plus JSON metadata: `id`, `width`, `height`, `rating`, `general_tags`, and `character_tags`. The selected label space has **12,476 tags**: 9,225 general tags, 3,247 character tags, and 4 rating tags.
+
+## Model Zoo: `dbv4-full`
+
+The tables below focus only on the main `dbv4-full` model line. Metrics are copied from the corresponding model cards and use the test split reported there.
+
+<img src="./assets/dbv4-full-model-snapshot.png" alt="dbv4-full model performance and parameter snapshot" width="100%">
+
+### Top 5 By Macro F1
+
+| Rank | Model | Family | Params | Macro@Best F1 | Macro@0.40 F1 | Micro@0.40 F1 |
+| --- | --- | --- | --- | --- | --- | --- |
+| 1 | [convnextv2_huge.dbv4-full](https://huggingface.co/animetimm/convnextv2_huge.dbv4-full) | ConvNeXt | 692.6M | 0.611 | 0.580 | 0.697 |
+| 2 | [eva02_large_patch14_448.dbv4-full](https://huggingface.co/animetimm/eva02_large_patch14_448.dbv4-full) | EVA | 316.8M | 0.599 | 0.569 | 0.693 |
+| 3 | [caformer_b36.dbv4-full](https://huggingface.co/animetimm/caformer_b36.dbv4-full) | CAFormer | 134.0M | 0.581 | 0.546 | 0.689 |
+| 4 | [swinv2_base_window8_256.dbv4-full](https://huggingface.co/animetimm/swinv2_base_window8_256.dbv4-full) | SwinV2 | 99.7M | 0.575 | 0.541 | 0.683 |
+| 5 | [caformer_m36.dbv4-full](https://huggingface.co/animetimm/caformer_m36.dbv4-full) | CAFormer | 82.7M | 0.559 | 0.515 | 0.676 |
+
+### Representative Models By Backbone Family
+
+Each row is the best `dbv4-full` model currently published for that backbone family.
+
+| Family | Model | Params | Macro@Best F1 | Macro@0.40 F1 | Micro@0.40 F1 |
+| --- | --- | --- | --- | --- | --- |
+| ConvNeXt | [convnextv2_huge.dbv4-full](https://huggingface.co/animetimm/convnextv2_huge.dbv4-full) | 692.6M | 0.611 | 0.580 | 0.697 |
+| EVA | [eva02_large_patch14_448.dbv4-full](https://huggingface.co/animetimm/eva02_large_patch14_448.dbv4-full) | 316.8M | 0.599 | 0.569 | 0.693 |
+| CAFormer | [caformer_b36.dbv4-full](https://huggingface.co/animetimm/caformer_b36.dbv4-full) | 134.0M | 0.581 | 0.546 | 0.689 |
+| SwinV2 | [swinv2_base_window8_256.dbv4-full](https://huggingface.co/animetimm/swinv2_base_window8_256.dbv4-full) | 99.7M | 0.575 | 0.541 | 0.683 |
+| ViT | [vit_base_patch16_224.dbv4-full](https://huggingface.co/animetimm/vit_base_patch16_224.dbv4-full) | 95.8M | 0.540 | 0.500 | 0.664 |
+| MobileNetV4 | [mobilenetv4_conv_aa_large.dbv4-full](https://huggingface.co/animetimm/mobilenetv4_conv_aa_large.dbv4-full) | 47.3M | 0.511 | 0.458 | 0.641 |
+| MobileNetV3 | [mobilenetv3_large_150d.dbv4-full](https://huggingface.co/animetimm/mobilenetv3_large_150d.dbv4-full) | 29.3M | 0.462 | 0.400 | 0.605 |
+| MobileViT | [mobilevitv2_200.dbv4-full](https://huggingface.co/animetimm/mobilevitv2_200.dbv4-full) | 30.2M | 0.454 | 0.401 | 0.608 |
+| ResNet | [resnet152.dbv4-full](https://huggingface.co/animetimm/resnet152.dbv4-full) | 83.7M | 0.486 | 0.448 | 0.624 |
+
+## Try It
+
+- [`dbv4-full-playground`](https://huggingface.co/spaces/animetimm/dbv4-full-playground) - tag images with pretrained `dbv4-full` models.
+- [`dbv4-full-ranklist`](https://huggingface.co/spaces/animetimm/dbv4-full-ranklist) - compare the public `dbv4-full` model lineup.
+
+## Maintenance
+
+The source data and chart builder are stored in this Space repository so the organization card can be regenerated without guessing:
+
+- [`data/dbv4_full_models.csv`](./data/dbv4_full_models.csv) - checked-in metric table.
+- [`data/dbv4_full_dataset_summary.json`](./data/dbv4_full_dataset_summary.json) - checked-in featured dataset summary.
+- [`data/featured_models.json`](./data/featured_models.json) - top-5 and best-by-family selections.
+- [`scripts/build_org_card.py`](./scripts/build_org_card.py) - regenerates the banner and model snapshot chart from the checked-in data.
+
+## Notes
+
+These releases are research and hobbyist infrastructure for visual tagging. Please check each model or dataset card for license, source data notes, intended use, and audience restrictions before reuse.

data/dbv4_full_dataset_summary.json

@@ -0,0 +1,41 @@
+{
+  "repo_id": "animetimm/danbooru-wdtagger-v4-w640-ws-full",
+  "pretrained_tag": "dbv4-full",
+  "source": "Danbooru-derived metadata cleaned by SmilingWolf, as described in the dataset card.",
+  "format": "WebDataset",
+  "image_rule": "Images are resized so min(width, height) <= 640.",
+  "splits": [
+    {
+      "name": "train",
+      "images": 5321713,
+      "total_size": "318 GB"
+    },
+    {
+      "name": "test",
+      "images": 295926,
+      "total_size": "17.7 GB"
+    },
+    {
+      "name": "val",
+      "images": 296957,
+      "total_size": "17.8 GB"
+    }
+  ],
+  "total_images": 5914596,
+  "total_size": "353.5 GB",
+  "sample_fields": [
+    "webp",
+    "json.id",
+    "json.width",
+    "json.height",
+    "json.rating",
+    "json.general_tags",
+    "json.character_tags"
+  ],
+  "selected_tags": {
+    "total": 12476,
+    "general": 9225,
+    "character": 3247,
+    "rating": 4
+  }
+}

data/dbv4_full_models.csv

@@ -0,0 +1,21 @@
+rank_by_macro_best,family,model,repo_id,params_m,macro_best,macro_040,micro_040
+1,ConvNeXt,convnextv2_huge.dbv4-full,animetimm/convnextv2_huge.dbv4-full,692.6,0.611,0.58,0.697
+2,EVA,eva02_large_patch14_448.dbv4-full,animetimm/eva02_large_patch14_448.dbv4-full,316.8,0.599,0.569,0.693
+3,CAFormer,caformer_b36.dbv4-full,animetimm/caformer_b36.dbv4-full,134.0,0.581,0.546,0.689
+4,SwinV2,swinv2_base_window8_256.dbv4-full,animetimm/swinv2_base_window8_256.dbv4-full,99.7,0.575,0.541,0.683
+5,CAFormer,caformer_m36.dbv4-full,animetimm/caformer_m36.dbv4-full,82.7,0.559,0.515,0.676
+6,ConvNeXt,convnext_base.dbv4-full,animetimm/convnext_base.dbv4-full,100.4,0.553,0.521,0.669
+7,CAFormer,caformer_s36.dbv4-full,animetimm/caformer_s36.dbv4-full,62.8,0.549,0.504,0.67
+8,ViT,vit_base_patch16_224.dbv4-full,animetimm/vit_base_patch16_224.dbv4-full,95.8,0.54,0.5,0.664
+9,CAFormer,caformer_s18.dbv4-full,animetimm/caformer_s18.dbv4-full,49.9,0.516,0.464,0.647
+10,MobileNetV4,mobilenetv4_conv_aa_large.dbv4-full,animetimm/mobilenetv4_conv_aa_large.dbv4-full,47.3,0.511,0.458,0.641
+11,ResNet,resnet152.dbv4-full,animetimm/resnet152.dbv4-full,83.7,0.486,0.448,0.624
+12,ResNet,resnet101.dbv4-full,animetimm/resnet101.dbv4-full,68.1,0.481,0.437,0.622
+13,MobileNetV3,mobilenetv3_large_150d.dbv4-full,animetimm/mobilenetv3_large_150d.dbv4-full,29.3,0.462,0.4,0.605
+14,MobileViT,mobilevitv2_200.dbv4-full,animetimm/mobilevitv2_200.dbv4-full,30.2,0.454,0.401,0.608
+15,ResNet,resnet50.dbv4-full,animetimm/resnet50.dbv4-full,49.1,0.421,0.382,0.572
+16,MobileNetV3,mobilenetv3_large_100.dbv4-full,animetimm/mobilenetv3_large_100.dbv4-full,20.2,0.383,0.311,0.533
+17,MobileNetV4,mobilenetv4_conv_small.dbv4-full,animetimm/mobilenetv4_conv_small.dbv4-full,18.5,0.379,0.312,0.529
+18,ResNet,resnet34.dbv4-full,animetimm/resnet34.dbv4-full,27.7,0.342,0.29,0.53
+19,ResNet,resnet18.dbv4-full,animetimm/resnet18.dbv4-full,17.6,0.314,0.259,0.499
+20,MobileNetV4,mobilenetv4_conv_small_050.dbv4-full,animetimm/mobilenetv4_conv_small_050.dbv4-full,16.9,0.3,0.194,0.429

data/dbv4_full_models.json

@@ -0,0 +1,202 @@
+[
+  {
+    "rank_by_macro_best": 1,
+    "family": "ConvNeXt",
+    "model": "convnextv2_huge.dbv4-full",
+    "repo_id": "animetimm/convnextv2_huge.dbv4-full",
+    "params_m": 692.6,
+    "macro_best": 0.611,
+    "macro_040": 0.58,
+    "micro_040": 0.697
+  },
+  {
+    "rank_by_macro_best": 2,
+    "family": "EVA",
+    "model": "eva02_large_patch14_448.dbv4-full",
+    "repo_id": "animetimm/eva02_large_patch14_448.dbv4-full",
+    "params_m": 316.8,
+    "macro_best": 0.599,
+    "macro_040": 0.569,
+    "micro_040": 0.693
+  },
+  {
+    "rank_by_macro_best": 3,
+    "family": "CAFormer",
+    "model": "caformer_b36.dbv4-full",
+    "repo_id": "animetimm/caformer_b36.dbv4-full",
+    "params_m": 134.0,
+    "macro_best": 0.581,
+    "macro_040": 0.546,
+    "micro_040": 0.689
+  },
+  {
+    "rank_by_macro_best": 4,
+    "family": "SwinV2",
+    "model": "swinv2_base_window8_256.dbv4-full",
+    "repo_id": "animetimm/swinv2_base_window8_256.dbv4-full",
+    "params_m": 99.7,
+    "macro_best": 0.575,
+    "macro_040": 0.541,
+    "micro_040": 0.683
+  },
+  {
+    "rank_by_macro_best": 5,
+    "family": "CAFormer",
+    "model": "caformer_m36.dbv4-full",
+    "repo_id": "animetimm/caformer_m36.dbv4-full",
+    "params_m": 82.7,
+    "macro_best": 0.559,
+    "macro_040": 0.515,
+    "micro_040": 0.676
+  },
+  {
+    "rank_by_macro_best": 6,
+    "family": "ConvNeXt",
+    "model": "convnext_base.dbv4-full",
+    "repo_id": "animetimm/convnext_base.dbv4-full",
+    "params_m": 100.4,
+    "macro_best": 0.553,
+    "macro_040": 0.521,
+    "micro_040": 0.669
+  },
+  {
+    "rank_by_macro_best": 7,
+    "family": "CAFormer",
+    "model": "caformer_s36.dbv4-full",
+    "repo_id": "animetimm/caformer_s36.dbv4-full",
+    "params_m": 62.8,
+    "macro_best": 0.549,
+    "macro_040": 0.504,
+    "micro_040": 0.67
+  },
+  {
+    "rank_by_macro_best": 8,
+    "family": "ViT",
+    "model": "vit_base_patch16_224.dbv4-full",
+    "repo_id": "animetimm/vit_base_patch16_224.dbv4-full",
+    "params_m": 95.8,
+    "macro_best": 0.54,
+    "macro_040": 0.5,
+    "micro_040": 0.664
+  },
+  {
+    "rank_by_macro_best": 9,
+    "family": "CAFormer",
+    "model": "caformer_s18.dbv4-full",
+    "repo_id": "animetimm/caformer_s18.dbv4-full",
+    "params_m": 49.9,
+    "macro_best": 0.516,
+    "macro_040": 0.464,
+    "micro_040": 0.647
+  },
+  {
+    "rank_by_macro_best": 10,
+    "family": "MobileNetV4",
+    "model": "mobilenetv4_conv_aa_large.dbv4-full",
+    "repo_id": "animetimm/mobilenetv4_conv_aa_large.dbv4-full",
+    "params_m": 47.3,
+    "macro_best": 0.511,
+    "macro_040": 0.458,
+    "micro_040": 0.641
+  },
+  {
+    "rank_by_macro_best": 11,
+    "family": "ResNet",
+    "model": "resnet152.dbv4-full",
+    "repo_id": "animetimm/resnet152.dbv4-full",
+    "params_m": 83.7,
+    "macro_best": 0.486,
+    "macro_040": 0.448,
+    "micro_040": 0.624
+  },
+  {
+    "rank_by_macro_best": 12,
+    "family": "ResNet",
+    "model": "resnet101.dbv4-full",
+    "repo_id": "animetimm/resnet101.dbv4-full",
+    "params_m": 68.1,
+    "macro_best": 0.481,
+    "macro_040": 0.437,
+    "micro_040": 0.622
+  },
+  {
+    "rank_by_macro_best": 13,
+    "family": "MobileNetV3",
+    "model": "mobilenetv3_large_150d.dbv4-full",
+    "repo_id": "animetimm/mobilenetv3_large_150d.dbv4-full",
+    "params_m": 29.3,
+    "macro_best": 0.462,
+    "macro_040": 0.4,
+    "micro_040": 0.605
+  },
+  {
+    "rank_by_macro_best": 14,
+    "family": "MobileViT",
+    "model": "mobilevitv2_200.dbv4-full",
+    "repo_id": "animetimm/mobilevitv2_200.dbv4-full",
+    "params_m": 30.2,
+    "macro_best": 0.454,
+    "macro_040": 0.401,
+    "micro_040": 0.608
+  },
+  {
+    "rank_by_macro_best": 15,
+    "family": "ResNet",
+    "model": "resnet50.dbv4-full",
+    "repo_id": "animetimm/resnet50.dbv4-full",
+    "params_m": 49.1,
+    "macro_best": 0.421,
+    "macro_040": 0.382,
+    "micro_040": 0.572
+  },
+  {
+    "rank_by_macro_best": 16,
+    "family": "MobileNetV3",
+    "model": "mobilenetv3_large_100.dbv4-full",
+    "repo_id": "animetimm/mobilenetv3_large_100.dbv4-full",
+    "params_m": 20.2,
+    "macro_best": 0.383,
+    "macro_040": 0.311,
+    "micro_040": 0.533
+  },
+  {
+    "rank_by_macro_best": 17,
+    "family": "MobileNetV4",
+    "model": "mobilenetv4_conv_small.dbv4-full",
+    "repo_id": "animetimm/mobilenetv4_conv_small.dbv4-full",
+    "params_m": 18.5,
+    "macro_best": 0.379,
+    "macro_040": 0.312,
+    "micro_040": 0.529
+  },
+  {
+    "rank_by_macro_best": 18,
+    "family": "ResNet",
+    "model": "resnet34.dbv4-full",
+    "repo_id": "animetimm/resnet34.dbv4-full",
+    "params_m": 27.7,
+    "macro_best": 0.342,
+    "macro_040": 0.29,
+    "micro_040": 0.53
+  },
+  {
+    "rank_by_macro_best": 19,
+    "family": "ResNet",
+    "model": "resnet18.dbv4-full",
+    "repo_id": "animetimm/resnet18.dbv4-full",
+    "params_m": 17.6,
+    "macro_best": 0.314,
+    "macro_040": 0.259,
+    "micro_040": 0.499
+  },
+  {
+    "rank_by_macro_best": 20,
+    "family": "MobileNetV4",
+    "model": "mobilenetv4_conv_small_050.dbv4-full",
+    "repo_id": "animetimm/mobilenetv4_conv_small_050.dbv4-full",
+    "params_m": 16.9,
+    "macro_best": 0.3,
+    "macro_040": 0.194,
+    "micro_040": 0.429
+  }
+]

data/featured_models.json

@@ -0,0 +1,146 @@
+{
+  "top5_macro_f1": [
+    {
+      "rank_by_macro_best": 1,
+      "family": "ConvNeXt",
+      "model": "convnextv2_huge.dbv4-full",
+      "repo_id": "animetimm/convnextv2_huge.dbv4-full",
+      "params_m": 692.6,
+      "macro_best": 0.611,
+      "macro_040": 0.58,
+      "micro_040": 0.697
+    },
+    {
+      "rank_by_macro_best": 2,
+      "family": "EVA",
+      "model": "eva02_large_patch14_448.dbv4-full",
+      "repo_id": "animetimm/eva02_large_patch14_448.dbv4-full",
+      "params_m": 316.8,
+      "macro_best": 0.599,
+      "macro_040": 0.569,
+      "micro_040": 0.693
+    },
+    {
+      "rank_by_macro_best": 3,
+      "family": "CAFormer",
+      "model": "caformer_b36.dbv4-full",
+      "repo_id": "animetimm/caformer_b36.dbv4-full",
+      "params_m": 134.0,
+      "macro_best": 0.581,
+      "macro_040": 0.546,
+      "micro_040": 0.689
+    },
+    {
+      "rank_by_macro_best": 4,
+      "family": "SwinV2",
+      "model": "swinv2_base_window8_256.dbv4-full",
+      "repo_id": "animetimm/swinv2_base_window8_256.dbv4-full",
+      "params_m": 99.7,
+      "macro_best": 0.575,
+      "macro_040": 0.541,
+      "micro_040": 0.683
+    },
+    {
+      "rank_by_macro_best": 5,
+      "family": "CAFormer",
+      "model": "caformer_m36.dbv4-full",
+      "repo_id": "animetimm/caformer_m36.dbv4-full",
+      "params_m": 82.7,
+      "macro_best": 0.559,
+      "macro_040": 0.515,
+      "micro_040": 0.676
+    }
+  ],
+  "best_by_family": [
+    {
+      "rank_by_macro_best": 1,
+      "family": "ConvNeXt",
+      "model": "convnextv2_huge.dbv4-full",
+      "repo_id": "animetimm/convnextv2_huge.dbv4-full",
+      "params_m": 692.6,
+      "macro_best": 0.611,
+      "macro_040": 0.58,
+      "micro_040": 0.697
+    },
+    {
+      "rank_by_macro_best": 2,
+      "family": "EVA",
+      "model": "eva02_large_patch14_448.dbv4-full",
+      "repo_id": "animetimm/eva02_large_patch14_448.dbv4-full",
+      "params_m": 316.8,
+      "macro_best": 0.599,
+      "macro_040": 0.569,
+      "micro_040": 0.693
+    },
+    {
+      "rank_by_macro_best": 3,
+      "family": "CAFormer",
+      "model": "caformer_b36.dbv4-full",
+      "repo_id": "animetimm/caformer_b36.dbv4-full",
+      "params_m": 134.0,
+      "macro_best": 0.581,
+      "macro_040": 0.546,
+      "micro_040": 0.689
+    },
+    {
+      "rank_by_macro_best": 4,
+      "family": "SwinV2",
+      "model": "swinv2_base_window8_256.dbv4-full",
+      "repo_id": "animetimm/swinv2_base_window8_256.dbv4-full",
+      "params_m": 99.7,
+      "macro_best": 0.575,
+      "macro_040": 0.541,
+      "micro_040": 0.683
+    },
+    {
+      "rank_by_macro_best": 8,
+      "family": "ViT",
+      "model": "vit_base_patch16_224.dbv4-full",
+      "repo_id": "animetimm/vit_base_patch16_224.dbv4-full",
+      "params_m": 95.8,
+      "macro_best": 0.54,
+      "macro_040": 0.5,
+      "micro_040": 0.664
+    },
+    {
+      "rank_by_macro_best": 10,
+      "family": "MobileNetV4",
+      "model": "mobilenetv4_conv_aa_large.dbv4-full",
+      "repo_id": "animetimm/mobilenetv4_conv_aa_large.dbv4-full",
+      "params_m": 47.3,
+      "macro_best": 0.511,
+      "macro_040": 0.458,
+      "micro_040": 0.641
+    },
+    {
+      "rank_by_macro_best": 13,
+      "family": "MobileNetV3",
+      "model": "mobilenetv3_large_150d.dbv4-full",
+      "repo_id": "animetimm/mobilenetv3_large_150d.dbv4-full",
+      "params_m": 29.3,
+      "macro_best": 0.462,
+      "macro_040": 0.4,
+      "micro_040": 0.605
+    },
+    {
+      "rank_by_macro_best": 14,
+      "family": "MobileViT",
+      "model": "mobilevitv2_200.dbv4-full",
+      "repo_id": "animetimm/mobilevitv2_200.dbv4-full",
+      "params_m": 30.2,
+      "macro_best": 0.454,
+      "macro_040": 0.401,
+      "micro_040": 0.608
+    },
+    {
+      "rank_by_macro_best": 11,
+      "family": "ResNet",
+      "model": "resnet152.dbv4-full",
+      "repo_id": "animetimm/resnet152.dbv4-full",
+      "params_m": 83.7,
+      "macro_best": 0.486,
+      "macro_040": 0.448,
+      "micro_040": 0.624
+    }
+  ]
+}

scripts/build_org_card.py

@@ -0,0 +1,124 @@
+#!/usr/bin/env python3
+"""Build the AnimeTimm organization card assets from checked-in model data."""
+
+import csv
+import json
+from collections import OrderedDict
+from pathlib import Path
+
+import matplotlib
+matplotlib.use("Agg")
+import matplotlib.pyplot as plt
+from matplotlib.patches import Rectangle
+
+ROOT = Path(__file__).resolve().parents[1]
+ASSETS = ROOT / "assets"
+DATA = ROOT / "data"
+ASSETS.mkdir(exist_ok=True)
+
+FAMILY_ORDER = [
+    "ConvNeXt",
+    "EVA",
+    "CAFormer",
+    "SwinV2",
+    "ViT",
+    "MobileNetV4",
+    "MobileNetV3",
+    "MobileViT",
+    "ResNet",
+]
+
+
+def load_rows():
+    with (DATA / "dbv4_full_models.csv").open(newline="", encoding="utf-8") as f:
+        rows = list(csv.DictReader(f))
+    for row in rows:
+        row["rank_by_macro_best"] = int(row["rank_by_macro_best"])
+        for key in ["params_m", "macro_best", "macro_040", "micro_040"]:
+            row[key] = float(row[key])
+    return sorted(rows, key=lambda item: item["macro_best"], reverse=True)
+
+
+def write_data_views(rows):
+    with (DATA / "dbv4_full_models.json").open("w", encoding="utf-8") as f:
+        json.dump(rows, f, indent=2)
+    best = OrderedDict()
+    for row in rows:
+        best.setdefault(row["family"], row)
+    family_rows = [best[family] for family in FAMILY_ORDER if family in best]
+    with (DATA / "featured_models.json").open("w", encoding="utf-8") as f:
+        json.dump({"top5_macro_f1": rows[:5], "best_by_family": family_rows}, f, indent=2)
+    return rows[:5], family_rows
+
+
+def draw_banner():
+    fig, ax = plt.subplots(figsize=(12, 3.6), dpi=160)
+    fig.patch.set_facecolor("#12131c")
+    ax.set_facecolor("#12131c")
+    ax.set_xlim(0, 1200)
+    ax.set_ylim(0, 360)
+    ax.axis("off")
+    colors = ["#ff7ab6", "#8f7cff", "#55d6be", "#ffd166", "#7bdff2"]
+    for i, color in enumerate(colors):
+        ax.add_patch(Rectangle((i * 240, 0), 240, 360, color=color, alpha=0.10 + 0.02 * i, lw=0))
+    for x in range(0, 1201, 60):
+        ax.plot([x, x], [0, 360], color="white", alpha=0.045, lw=0.7)
+    for y in range(0, 361, 60):
+        ax.plot([0, 1200], [y, y], color="white", alpha=0.045, lw=0.7)
+    for x, y, size, color in [(930, 240, 52, "#ff7ab6"), (1010, 165, 38, "#55d6be"), (1080, 260, 28, "#ffd166"), (845, 120, 32, "#8f7cff")]:
+        ax.scatter([x], [y], s=size * size / 4, marker="*", color=color, alpha=0.85, edgecolors="white", linewidths=0.8)
+    labels = ["tags", "timm", "anime", "F1", "ONNX", "safetensors"]
+    positions = [(720, 78), (812, 285), (935, 65), (1045, 104), (705, 210), (1085, 205)]
+    for (x, y), label, color in zip(positions, labels, colors + ["#ffffff"]):
+        ax.text(x, y, label, color="#f7f7fb", fontsize=13, fontweight="bold", ha="center", va="center", bbox=dict(boxstyle="round,pad=0.35,rounding_size=0.18", fc=color, ec="white", alpha=0.35, lw=0.8))
+    ax.text(70, 218, "AnimeTimm", color="white", fontsize=54, fontweight="bold", ha="left", va="center")
+    ax.text(75, 157, "timm-based vision models for anime-style image tagging", color="#e6e7f3", fontsize=20, ha="left", va="center")
+    ax.text(76, 108, "A DeepGHS research-and-hobbyist project", color="#ffcfdf", fontsize=15, ha="left", va="center")
+    xs = [720, 800, 890, 980, 1080]
+    ys = [135, 175, 205, 225, 252]
+    ax.plot(xs, ys, color="#ffffff", lw=2.2, alpha=0.78)
+    ax.scatter(xs, ys, s=75, color="#55d6be", edgecolor="white", linewidth=1.2, zorder=3)
+    fig.savefig(ASSETS / "animetimm-banner.png", bbox_inches="tight", pad_inches=0)
+    plt.close(fig)
+
+
+def draw_snapshot(top5, family_rows):
+    plot_rows = top5 + [row for row in family_rows if row not in top5]
+    plot_rows = sorted(plot_rows, key=lambda row: row["macro_best"])
+    fig, ax1 = plt.subplots(figsize=(11.5, 7.2), dpi=160)
+    fig.patch.set_facecolor("#ffffff")
+    ax1.set_facecolor("#fbfbff")
+    y = list(range(len(plot_rows)))
+    labels = [row["model"].replace(".dbv4-full", "") for row in plot_rows]
+    bar_colors = ["#8f7cff" if row in top5 else "#55b9d6" for row in plot_rows]
+    ax1.barh(y, [row["macro_best"] for row in plot_rows], color=bar_colors, alpha=0.88)
+    ax1.set_yticks(y)
+    ax1.set_yticklabels(labels, fontsize=9)
+    ax1.set_xlim(0.28, 0.63)
+    ax1.set_xlabel("Macro@Best F1", fontsize=11)
+    ax1.grid(axis="x", alpha=0.22)
+    for idx, row in enumerate(plot_rows):
+        ax1.text(row["macro_best"] + 0.004, idx, f"{row['macro_best']:.3f}", va="center", fontsize=8, color="#222222")
+    ax2 = ax1.twiny()
+    ax2.scatter([row["params_m"] for row in plot_rows], y, color="#ff7ab6", s=46, edgecolor="white", linewidth=0.8, zorder=5)
+    ax2.set_xscale("log")
+    ax2.set_xlabel("Parameters (M, log scale)", fontsize=11)
+    ax2.set_xlim(12, 900)
+    for idx, row in enumerate(plot_rows):
+        ax2.text(row["params_m"] * 1.04, idx + 0.14, f"{row['params_m']:.1f}M", va="center", fontsize=7, color="#8a1f55")
+    ax1.set_title("AnimeTimm dbv4-full Model Snapshot", fontsize=16, fontweight="bold", pad=14)
+    fig.text(0.125, 0.028, "Purple bars mark the top-5 Macro@Best F1 models. Blue bars are best-per-family representatives not already in top-5.", fontsize=8.5, color="#555555")
+    fig.tight_layout(rect=[0, 0.045, 1, 1])
+    fig.savefig(ASSETS / "dbv4-full-model-snapshot.png", bbox_inches="tight")
+    plt.close(fig)
+
+
+def main():
+    rows = load_rows()
+    top5, family_rows = write_data_views(rows)
+    draw_banner()
+    draw_snapshot(top5, family_rows)
+
+
+if __name__ == "__main__":
+    main()