Merge dev into main

.gitattributes

@@ -33,3 +33,5 @@ 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/demo_banner.mp4 filter=lfs diff=lfs merge=lfs -text
+assets/demo_banner_poster.jpg filter=lfs diff=lfs merge=lfs -text

README.md

@@ -0,0 +1,250 @@
+---
+license: apache-2.0
+
+pipeline_tag: image-classification
+
+library_name: timm
+
+datasets:
+  - dronefreak/BDD100K-Weather-Classification
+
+tags:
+  - image-classification
+  - bdd100k-toolkit
+  - timm
+  - pytorch
+  - computer-vision
+  - autonomous-driving
+  - self-driving-cars
+  - driving-scenes
+  - weather-classification
+metrics:
+  - accuracy
+  - f1
+  - precision
+  - recall
+
+base_model: "timm/efficientformerv2_l.snap_dist_in1k"
+
+model-index:
+  - name: EfficientFormerV2-L Finetuned on BDD100K Weather Classification
+    results:
+      - task:
+          type: image-classification
+          name: Image Classification
+        dataset:
+          name: BDD100K Weather Classification
+          type: dronefreak/BDD100K-Weather-Classification
+        metrics:
+          - type: accuracy
+            value: 83.27
+            name: Top-1 accuracy (test split)
+          - type: f1
+            value: 65.93
+            name: Macro F1 (test split)
+          - type: precision
+            value: 67.12
+            name: Macro precision (test split)
+          - type: recall
+            value: 65.03
+            name: Macro recall (test split)
+        source:
+          url: https://github.com/dronefreak/bdd100k-toolkit
+          name: BDD100K-Toolkit
+---
+
+# EfficientFormerV2-L Finetuned on BDD100K Weather Classification
+
+<p align="center"><video controls autoplay loop muted playsinline width="900" poster="https://huggingface.co/dronefreak/bdd100k-weather-efficientformerv2_l/resolve/main/assets/demo_banner_poster.jpg" src="https://huggingface.co/dronefreak/bdd100k-weather-efficientformerv2_l/resolve/main/assets/demo_banner.mp4"><img src="https://huggingface.co/dronefreak/bdd100k-weather-efficientformerv2_l/resolve/main/assets/demo_banner_poster.jpg" alt="EfficientFormerV2-L weather predictions on two BDD100K test clips" width="900"></video></p>
+
+<br>
+
+<!-- ROW 1: Identity & Tech Stack -->
+<div style="display: flex; justify-content: center; align-items: center; gap: 8px; margin-bottom: 8px; flex-wrap: wrap;">
+  <img src="https://img.shields.io/badge/Task-Image_Classification-blue?style=flat-square" alt="Task">
+  <img src="https://img.shields.io/badge/Framework-timm-0aa1a7?style=flat-square" alt="Framework">
+  <img src="https://img.shields.io/badge/Base_Model-EfficientFormerV2--L-purple?style=flat-square" alt="Base Model">
+</div>
+
+<!-- ROW 2: Performance Metrics -->
+<div style="display: flex; justify-content: center; align-items: center; gap: 8px; margin-bottom: 8px; flex-wrap: wrap;">
+  <img src="https://img.shields.io/badge/Macro_F1-65.93%25-success?style=flat-square" alt="Macro F1">
+  <img src="https://img.shields.io/badge/Top--1-83.27%25-orange?style=flat-square" alt="Top-1">
+  <img src="https://img.shields.io/badge/Params-25.6M-lightgrey?style=flat-square" alt="Params">
+</div>
+
+<!-- ROW 3: Metadata -->
+<div style="display: flex; justify-content: center; align-items: center; gap: 8px; margin-bottom: 24px; flex-wrap: wrap;">
+  <img src="https://img.shields.io/badge/License-Apache--2.0-lightgrey?style=flat-square" alt="License">
+  <a href="https://github.com/dronefreak/bdd100k-toolkit"><img src="https://img.shields.io/badge/Source-BDD100K--Toolkit-black?style=flat-square" alt="Source"></a>
+</div>
+
+Fine-tuned EfficientFormerV2-L image classifier on the **BDD100K Weather Classification** dataset, trained and evaluated as part of [BDD100K-Toolkit](https://github.com/dronefreak/bdd100k-toolkit), a dependency-clean toolkit for preparing BDD100K, training models on it and evaluating them with the same metrics on the same splits.
+
+7-class weather classification (clear / partly cloudy / overcast / rainy / snowy / foggy / unknown) derived from BDD100K's per-image attributes.weather field. Unofficial task; follows the Kaggle dataset of the same name.
+
+---
+
+## Usage
+
+```python
+import timm, torch
+from huggingface_hub import hf_hub_download
+from PIL import Image
+from torchvision import transforms as T
+
+ckpt = torch.load(
+    hf_hub_download("dronefreak/bdd100k-weather-efficientformerv2_l", "best.pt"), map_location="cpu", weights_only=True
+)
+model = timm.create_model(
+    ckpt["model_name"], pretrained=False, num_classes=len(ckpt["class_names"])
+)
+model.load_state_dict(ckpt["state_dict"])
+model.eval()
+prep = T.Compose(
+    [T.Resize((ckpt["imgsz"],) * 2), T.ToTensor(), T.Normalize(ckpt["mean"], ckpt["std"])]
+)
+with torch.no_grad():
+    probs = model(prep(Image.open("street.jpg").convert("RGB"))[None]).softmax(1)[0]
+print(ckpt["class_names"][probs.argmax()], f"{probs.max():.1%}")
+```
+
+---
+
+## Results
+
+Evaluated on the `test` split (10000 images).
+
+| Metric | Value |
+|---|---|
+| Macro F1 | 65.93% |
+| Balanced accuracy | 65.03% |
+| Macro precision | 67.12% |
+| Macro recall | 65.03% |
+| Top-1 accuracy | 83.27% |
+| Top-5 accuracy | 99.78% |
+
+### Per class
+
+| Class | Precision | Recall | F1 | Test images |
+|---|---|---|---|---|
+| clear | 91.15% | 92.63% | 91.88% | 5346 |
+| foggy | 0.00% | 0.00% | 0.00% | 13 (few) |
+| overcast | 66.38% | 69.49% | 67.90% | 1239 |
+| partly cloudy | 69.10% | 65.45% | 67.22% | 738 |
+| rainy | 84.83% | 70.46% | 76.98% | 738 |
+| snowy | 86.28% | 79.32% | 82.66% | 769 |
+| unknown | 72.08% | 77.87% | 74.86% | 1157 |
+
+This checkpoint classifies none of the test images of `foggy` (13 images) correctly, so treat it as unsupported.
+
+---
+
+## Model Zoo
+
+All runs below were evaluated on the same `test` split, sorted by top-1 accuracy.
+
+| Model | Top-1 | Macro F1 | Balanced acc | Macro precision |
+|---|---|---|---|---|
+| TinyViT-21M | 83.90% | 68.65% | 67.15% | 74.81% |
+| EfficientViT-B3 | 83.51% | 68.17% | 66.34% | 81.75% |
+| EfficientViT-B2 | 83.46% | 66.07% | 65.04% | 67.54% |
+| **EfficientFormerV2-L** | **83.27%** | **65.93%** | **65.03%** | **67.12%** |
+| RepViT-M1.5 | 83.19% | 67.78% | 65.71% | 81.56% |
+| EfficientViT-L1 | 83.13% | 65.70% | 65.11% | 66.46% |
+| RepViT-M2.3 | 83.02% | 65.46% | 64.33% | 67.02% |
+| ConvNeXt-Atto | 83.00% | 67.44% | 65.25% | 81.38% |
+| EfficientFormerV2-S2 | 82.90% | 65.50% | 64.68% | 66.98% |
+| EfficientViT-B0 | 82.79% | 65.10% | 63.62% | 67.11% |
+| EfficientViT-B1 | 82.67% | 64.95% | 63.59% | 66.83% |
+| MobileNetV4-Conv-Large | 82.20% | 66.62% | 64.68% | 80.26% |
+| ResNet-18 | 82.19% | 64.30% | 62.97% | 66.08% |
+| MobileNetV4-Conv-Small | 82.16% | 66.51% | 64.37% | 80.44% |
+| YOLO26n | 82.04% | 64.09% | 62.46% | 66.57% |
+| YOLO11s | 81.91% | 63.81% | 62.05% | 66.35% |
+| YOLO26s | 81.46% | 65.14% | 62.61% | 69.75% |
+| YOLOv8s | 81.40% | 66.34% | 63.52% | 71.54% |
+| YOLO11n | 81.32% | 62.91% | 61.19% | 65.68% |
+| YOLOv8n | 81.22% | 63.07% | 61.42% | 65.52% |
+
+---
+
+## Training
+
+| Setting | Value |
+|---|---|
+| Epochs (max) | 50 |
+| Epochs (trained) | 16 |
+| Best epoch (`best.pt`) | 6 |
+| `best.pt` chosen by | macro_f1 on the valid split |
+| Early stopping patience | 10 |
+| Batch size | 64 |
+| Image size | 224 |
+| Optimizer | auto, resolved to AdamW (peak lr 3e-04) |
+| Weights | EMA |
+
+---
+
+## Dataset
+
+[dronefreak/BDD100K-Weather-Classification](https://huggingface.co/datasets/dronefreak/BDD100K-Weather-Classification) holds the prepared splits these models were trained and evaluated on.
+
+---
+
+## Limitations
+
+- Unofficial task: labels are BDD100K's per-image attributes, not a benchmark with a public leaderboard, so scores are only comparable with other models evaluated on this split.
+- Not the official test set: `test` here is BDD100K's official validation split (the official test labels are not released) and `valid` is a seeded 15% slice of the official train split.
+- Heavily imbalanced: rare classes have very few test images, so their per-class scores are noisy. Prefer macro F1 over accuracy.
+- Images are US dashcam frames; generalization to other regions or camera setups is untested.
+- BDD100K is released for non-commercial research and education. Check its license before any use of these weights beyond research.
+- `clear`, `partly cloudy` and `overcast` are subjective annotator calls, so expect label noise between neighbouring classes.
+
+---
+
+## License
+
+The weights are released under the license in the metadata above. They were trained on BDD100K, which is free for non-commercial research and education; commercial use needs separate permission (see https://www.bdd100k.com/). The prepared dataset on the Hub is tagged `license: other`, and the original BDD100K terms still apply.
+
+---
+
+## Citation
+
+### Model
+
+```bibtex
+@article{li2022rethinking,
+  title={Rethinking Vision Transformers for MobileNet Size and Speed},
+  author={Li, Yanyu and Hu, Ju and Wen, Yang and Evangelidis, Georgios and Salahi, Kamyar and Wang, Yanzhi and Tulyakov, Sergey and Ren, Jian},
+  journal={arXiv preprint arXiv:2212.08059},
+  year={2022}
+}
+```
+
+### Dataset
+
+```bibtex
+@article{yu2018bdd100k,
+  title={BDD100K: A Diverse Driving Dataset for Heterogeneous Multitask Learning},
+  author={Yu, Fisher and Chen, Haofeng and Wang, Xin and Xian, Wenqi and Chen, Yingying and Liu, Fangchen and Madhavan, Vashisht and Darrell, Trevor},
+  journal={arXiv preprint arXiv:1805.04687},
+  year={2018}
+}
+```
+
+---
+
+## Files
+
+- `best.pt`
+- `metrics.json`
+- `results.csv`
+- `assets/demo_banner.mp4`
+- `assets/demo_banner_poster.jpg`
+- `README.md`
+
+---
+
+## Reproduce
+
+Trained and evaluated with [BDD100K-Toolkit](https://github.com/dronefreak/bdd100k-toolkit): `bdd100k-evaluate --dataset bdd100k-weather --checkpoint <weights> --data-dir <prepared dir> --split test`.

assets/demo_banner.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e363dce06dfe2a6c1fe1accf477e0f254decde0733e558133eab59bce8ac259d
+size 7926855

best.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e99d4ae6c8544ee5b75783465c327a7ccd2f92bdbf0e8eae936233270806e376
+size 103376093

metrics.json

@@ -0,0 +1,127 @@
+{
+  "accuracy": 0.8327,
+  "balanced_accuracy": 0.6503242813775042,
+  "macro_precision": 0.6711687705534296,
+  "macro_recall": 0.6503242813775042,
+  "macro_f1": 0.6592982465936582,
+  "per_class": {
+    "clear": {
+      "precision": 0.9114669611632615,
+      "recall": 0.9263000374111485,
+      "f1": 0.9188236385564523,
+      "support": 5346
+    },
+    "foggy": {
+      "precision": 0.0,
+      "recall": 0.0,
+      "f1": 0.0,
+      "support": 13
+    },
+    "overcast": {
+      "precision": 0.6638396299151889,
+      "recall": 0.6949152542372882,
+      "f1": 0.6790220820189274,
+      "support": 1239
+    },
+    "partly cloudy": {
+      "precision": 0.6909871244635193,
+      "recall": 0.6544715447154471,
+      "f1": 0.6722338204592903,
+      "support": 738
+    },
+    "rainy": {
+      "precision": 0.8482871125611745,
+      "recall": 0.7046070460704607,
+      "f1": 0.76980014803849,
+      "support": 738
+    },
+    "snowy": {
+      "precision": 0.8628005657708628,
+      "recall": 0.7932379713914174,
+      "f1": 0.8265582655826559,
+      "support": 769
+    },
+    "unknown": {
+      "precision": 0.7208,
+      "recall": 0.7787381158167676,
+      "f1": 0.7486497714997922,
+      "support": 1157
+    }
+  },
+  "top1_accuracy": 0.8327,
+  "top5_accuracy": 0.9978,
+  "class_names": [
+    "clear",
+    "foggy",
+    "overcast",
+    "partly cloudy",
+    "rainy",
+    "snowy",
+    "unknown"
+  ],
+  "confusion_matrix": [
+    [
+      4952,
+      0,
+      93,
+      118,
+      36,
+      50,
+      97
+    ],
+    [
+      8,
+      0,
+      1,
+      0,
+      3,
+      0,
+      1
+    ],
+    [
+      62,
+      1,
+      861,
+      85,
+      21,
+      7,
+      202
+    ],
+    [
+      85,
+      0,
+      154,
+      483,
+      2,
+      6,
+      8
+    ],
+    [
+      117,
+      0,
+      50,
+      3,
+      520,
+      22,
+      26
+    ],
+    [
+      95,
+      0,
+      24,
+      6,
+      19,
+      610,
+      15
+    ],
+    [
+      114,
+      0,
+      114,
+      4,
+      12,
+      12,
+      901
+    ]
+  ]
+}

results.csv

@@ -0,0 +1,17 @@
+epoch,train_loss,valid_loss,valid_accuracy,valid_balanced_accuracy,valid_macro_f1,monitored_macro_f1,lr,seconds
+1,0.97208,0.56976,0.81592,0.60686,0.62819,0.62819,3.000e-04,233.5
+2,0.58313,0.52103,0.82498,0.63578,0.64476,0.64476,2.997e-04,232.8
+3,0.52444,0.49826,0.82956,0.65017,0.66385,0.66385,2.988e-04,232.8
+4,0.49029,0.49662,0.83271,0.65103,0.66705,0.66705,2.972e-04,232.9
+5,0.45996,0.50734,0.83147,0.65109,0.67463,0.67463,2.951e-04,232.8
+6,0.42315,0.51467,0.83100,0.66130,0.67810,0.67810,2.924e-04,232.7
+7,0.37145,0.54582,0.82231,0.65392,0.66698,0.66698,2.890e-04,232.8
+8,0.31658,0.59350,0.82069,0.65022,0.66445,0.66445,2.851e-04,232.9
+9,0.26390,0.64890,0.81993,0.64817,0.66484,0.66484,2.807e-04,232.8
+10,0.21873,0.70386,0.81659,0.64360,0.66003,0.66003,2.757e-04,232.9
+11,0.18370,0.77018,0.81000,0.64325,0.66436,0.66436,2.702e-04,232.8
+12,0.15502,0.83533,0.81029,0.64437,0.66550,0.66550,2.642e-04,232.7
+13,0.13500,0.86741,0.80962,0.63231,0.64964,0.64964,2.578e-04,232.8
+14,0.12043,0.90606,0.81086,0.63947,0.65175,0.65175,2.508e-04,233.4
+15,0.10627,0.93525,0.81048,0.62955,0.64047,0.64047,2.435e-04,232.9
+16,0.09638,0.97081,0.81134,0.63698,0.65085,0.65085,2.358e-04,232.9