Upload folder using huggingface_hub

README.md

@@ -1,3 +1,57 @@
----
-license: mit
----
+---
+license: mit
+pipeline_tag: image-classification
+tags:
+  - onnx
+  - onnxruntime
+  - image-classification
+  - binary-classification
+library_name: onnx
+---
+
+# cstria0106/shit-detector
+
+ONNX image classifier for detecting whether an image contains feces or toilet waste/staining under this project's labeling policy.
+
+## Files
+
+- `shit_detector.onnx`: ONNX model exported with dynamic batch axis.
+- `metadata.json`: preprocessing values, threshold, class names, evaluation metrics, and deployment notes.
+- `inference.py`: minimal ONNX Runtime inference helper.
+- `requirements.txt`: runtime dependencies for the helper.
+
+## Labels
+
+Class order: `['shit', 'not_shit']`.
+
+The positive class is `shit`. The default decision threshold is `0.149` on the positive-class probability.
+
+## Evaluation Snapshot
+
+- ID test precision: 93.51%
+- ID test recall: 91.53%
+- ID test F1: 92.51%
+- ROC-AUC: 99.09%
+- PR-AUC: 97.42%
+- Policy-positive recall: 94.57%
+- Hard-negative guard false-positive rate: 3.95%
+
+These metrics come from the local evaluation set recorded in `metadata.json`; they are not a public benchmark.
+
+## Intended Use
+
+This model is intended for project-specific binary image triage. It may produce false positives on visually similar brown food, stains, mud, and other hard negatives, and false negatives on small, occluded, color-shifted, or unusual positive cases.
+
+## Runtime
+
+```bash
+pip install -r requirements.txt
+python inference.py path/to/image.jpg
+```
+
+## Deployment Target
+
+- Target: `lattepanda_n4120`
+- CPU: `Intel Celeron N4120`
+- Model parameters: `4010110`
+- ONNX size MB: `15.341004371643066`

config.json

@@ -0,0 +1,28 @@
+{
+  "architectures": [
+    "tf_efficientnet_b0.ns_jft_in1k"
+  ],
+  "model_type": "image-classification",
+  "library_name": "onnx",
+  "task": "image-classification",
+  "labels": [
+    "shit",
+    "not_shit"
+  ],
+  "positive_label": "shit",
+  "positive_threshold": 0.149,
+  "preprocessor": {
+    "input_size": 256,
+    "mean": [
+      0.485,
+      0.456,
+      0.406
+    ],
+    "std": [
+      0.229,
+      0.224,
+      0.225
+    ],
+    "resize": "shortest_edge_then_center_crop"
+  }
+}

inference.py

@@ -0,0 +1,63 @@
+from __future__ import annotations
+
+import json
+import sys
+from pathlib import Path
+
+import numpy as np
+import onnxruntime as ort
+from PIL import Image
+
+
+MODEL_DIR = Path(__file__).resolve().parent
+
+
+def _resize_center_crop(image: Image.Image, size: int) -> Image.Image:
+    resize_short_edge = int(size * 1.14)
+    width, height = image.size
+    scale = resize_short_edge / min(width, height)
+    resized = image.resize((round(width * scale), round(height * scale)))
+    left = (resized.width - size) // 2
+    top = (resized.height - size) // 2
+    return resized.crop((left, top, left + size, top + size))
+
+
+def _preprocess(image: Image.Image, metadata: dict) -> np.ndarray:
+    input_size = int(metadata["input_size"])
+    image = _resize_center_crop(image.convert("RGB"), input_size)
+    array = np.asarray(image, dtype=np.float32) / 255.0
+    array = np.transpose(array, (2, 0, 1))
+    mean = np.asarray(metadata["mean"], dtype=np.float32)[:, None, None]
+    std = np.asarray(metadata["std"], dtype=np.float32)[:, None, None]
+    return ((array - mean) / std)[None, ...]
+
+
+def predict(image_path: str | Path) -> dict:
+    metadata = json.loads((MODEL_DIR / "metadata.json").read_text(encoding="utf-8"))
+    session = ort.InferenceSession(
+        str(MODEL_DIR / "shit_detector.onnx"),
+        providers=["CPUExecutionProvider"],
+    )
+    batch = _preprocess(Image.open(image_path), metadata)
+    logits = session.run(None, {session.get_inputs()[0].name: batch})[0]
+    logits = logits * float(metadata.get("logit_scale", 1.0))
+    shifted = logits - logits.max(axis=-1, keepdims=True)
+    probs = np.exp(shifted) / np.exp(shifted).sum(axis=-1, keepdims=True)
+    shit_probability = float(probs[0, 0])
+    confidence = float(probs.max(axis=-1)[0])
+    threshold = float(metadata.get("shit_threshold", 0.5))
+    label = "shit" if shit_probability >= threshold else "not_shit"
+    return {
+        "status": label,
+        "label": label,
+        "is_shit": label == "shit",
+        "confidence": confidence,
+        "shit_probability": shit_probability,
+        "threshold": threshold,
+    }
+
+
+if __name__ == "__main__":
+    if len(sys.argv) != 2:
+        raise SystemExit("Usage: python inference.py path/to/image")
+    print(json.dumps(predict(sys.argv[1]), indent=2))

metadata.json

@@ -0,0 +1,160 @@
+{
+  "model_name": "tf_efficientnet_b0.ns_jft_in1k",
+  "input_size": 256,
+  "mean": [
+    0.485,
+    0.456,
+    0.406
+  ],
+  "std": [
+    0.229,
+    0.224,
+    0.225
+  ],
+  "class_names": [
+    "shit",
+    "not_shit"
+  ],
+  "model_params": 4010110,
+  "deployment": {
+    "target_name": "lattepanda_n4120",
+    "cpu": "Intel Celeron N4120",
+    "cpu_threads": 4,
+    "gpu": "Intel UHD Graphics 600",
+    "memory_gib": 3.66,
+    "max_model_params": 6000000,
+    "max_onnx_size_mb": 32,
+    "max_cpu_latency_ms_p95": 1500,
+    "max_process_rss_mb": 1024
+  },
+  "shit_threshold": 0.149,
+  "f1_shit_threshold": 0.149,
+  "logit_scale": 1.0,
+  "threshold_policy": "shit_threshold is selected on validation F1.",
+  "quality_targets": {
+    "precision": 0.99,
+    "recall": 0.8,
+    "f1": 0.9,
+    "roc_auc": 0.95,
+    "pr_auc": 0.95
+  },
+  "delete_allowed_negative_fragments": [],
+  "last_eval": {
+    "val_f1_threshold": {
+      "precision": 0.9727272727272728,
+      "recall": 0.8916666666666667,
+      "f1": 0.9304347826086957,
+      "false_positive_rate": 0.004373177842565598,
+      "tp": 107,
+      "shit_tp": 107,
+      "delete_allowed_tp": 0,
+      "fp": 3,
+      "fn": 13,
+      "tn": 683,
+      "delete_allowed_count": 0,
+      "delete_allowed_predicted_shit_count": 0
+    },
+    "id_test_at_val_threshold": {
+      "precision": 0.9351351351351351,
+      "recall": 0.9153439153439153,
+      "f1": 0.9251336898395722,
+      "false_positive_rate": 0.011964107676969093,
+      "tp": 173,
+      "shit_tp": 173,
+      "delete_allowed_tp": 0,
+      "fp": 12,
+      "fn": 16,
+      "tn": 991,
+      "delete_allowed_count": 0,
+      "delete_allowed_predicted_shit_count": 0
+    },
+    "id_test_oracle_f1": {
+      "precision": 0.9558011049723757,
+      "recall": 0.9153439153439153,
+      "f1": 0.9351351351351351,
+      "false_positive_rate": 0.007976071784646061,
+      "tp": 173,
+      "shit_tp": 173,
+      "delete_allowed_tp": 0,
+      "fp": 8,
+      "fn": 16,
+      "tn": 995,
+      "delete_allowed_count": 0,
+      "delete_allowed_predicted_shit_count": 0
+    },
+    "roc_auc": 0.9909029525181069,
+    "pr_auc": 0.9741673091947349,
+    "hard_negative_guard": {
+      "false_shit_rate": 0.03954802259887006,
+      "false_shit_count": 14,
+      "not_shit_count": 340,
+      "total": 354
+    },
+    "policy_positive": {
+      "recall": 0.9456521739130435,
+      "not_shit_miss_count": 20,
+      "total": 368,
+      "by_group": {
+        "brown_water_floating": {
+          "total": 28,
+          "predicted_shit_count": 27,
+          "recall": 0.9642857142857143
+        },
+        "clogged_toilet": {
+          "total": 19,
+          "predicted_shit_count": 17,
+          "recall": 0.8947368421052632
+        },
+        "color_shifted": {
+          "total": 20,
+          "predicted_shit_count": 19,
+          "recall": 0.95
+        },
+        "manual_new_shits": {
+          "total": 6,
+          "predicted_shit_count": 6,
+          "recall": 1.0
+        },
+        "other_shit": {
+          "total": 29,
+          "predicted_shit_count": 26,
+          "recall": 0.896551724137931
+        },
+        "subagent_fp_policy_positive": {
+          "total": 2,
+          "predicted_shit_count": 2,
+          "recall": 1.0
+        },
+        "toilet_feces_general": {
+          "total": 171,
+          "predicted_shit_count": 164,
+          "recall": 0.9590643274853801
+        },
+        "toilet_soiling": {
+          "total": 50,
+          "predicted_shit_count": 44,
+          "recall": 0.88
+        },
+        "toilet_urine_soiling": {
+          "total": 43,
+          "predicted_shit_count": 43,
+          "recall": 1.0
+        }
+      },
+      "min_group_recall": 0.88
+    }
+  },
+  "onnx": {
+    "path": "models\\shit_detector.onnx",
+    "size_mb": 15.341004371643066,
+    "opset_version": 17,
+    "providers": [
+      "CPUExecutionProvider"
+    ]
+  },
+  "checkpoint": {
+    "path": "checkpoints\\best.pt",
+    "selection_metric": "shit_f1",
+    "selection_score": 0.9304347826086957
+  }
+}

requirements.txt

@@ -0,0 +1,3 @@
+numpy>=1.26
+onnxruntime>=1.26
+pillow>=10.0

shit_detector.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ffed6c44cbeb8d973f8c91daa04dc70da5fd638eb27fc4add50f425791d9056a
+size 16086209