Merge pull request #37 from dronefreak/hac-video-ucf

Hac video classificaiton ucf101

Author: dronefreak

.github/README.md

@@ -4,6 +4,7 @@
 [![🔥 PyTorch 2.0+](https://img.shields.io/badge/PyTorch-2.0+-EE4C2C?logo=pytorch&logoColor=white&style=for-the-badge)](https://pytorch.org/)
 [![🤗 HuggingFace Models](https://img.shields.io/badge/HuggingFace-Models-FFD21E?logo=huggingface&logoColor=yellow&style=for-the-badge)](https://huggingface.co/dronefreak/human-action-classification-stanford40)
 [![⚖️ License: Apache 2.0](https://img.shields.io/badge/License-Apache%202.0-green?logo=open-source-initiative&logoColor=white&style=for-the-badge)](https://opensource.org/licenses/Apache-2.0)
+[![⭐ GitHub stars](https://img.shields.io/github/stars/dronefreak/human-action-classification?style=for-the-badge)](https://github.com/dronefreak/human-action-classification/stargazers)
 
 **The fastest way to classify human actions in images and video.** MediaPipe pose estimation + PyTorch CNNs with **90 FPS real-time performance**. Zero C++ compilation, pure Python.
 

src/hac/__init__.py

@@ -1,13 +1,21 @@
-"""Human Action Classification v2.0.
-
-Modern action recognition using MediaPipe pose estimation and PyTorch. Completely
-rewritten from the TensorFlow 1.13 version.
-"""
+"""Human Action Classification - Image and Video."""
 
 __version__ = "2.0.0"
-__author__ = "Saumya Kumaar Saksena"
 
-from hac.inference.predictor import ActionPredictor
-from hac.models.classifier import ActionClassifier
+# Common utilities
+from .common.metrics import compute_accuracy, compute_metrics
+
+# Image module (backward compatibility)
+from .image.inference.predictor import ActionPredictor as ImagePredictor
+from .image.models.classifier import ActionClassifier
+
+# Video module
+from .video.models.classifier import Video3DCNN
 
-__all__ = ["ActionPredictor", "ActionClassifier"]
+__all__ = [
+    "ImagePredictor",
+    "ActionClassifier",
+    "Video3DCNN",
+    "compute_accuracy",
+    "compute_metrics",
+]

src/hac/cli.py

@@ -3,7 +3,7 @@
 import argparse
 import sys
 
-from hac.inference.predictor import ActionPredictor
+from hac.image.inference.predictor import ActionPredictor
 
 
 def infer():

src/hac/common/__init__.py

@@ -0,0 +1,15 @@
+"""Common utilities shared between image and video modules."""
+
+from .convert_to_safetensors import convert_checkpoint
+from .transforms import (
+    denormalize_image,
+    get_inference_transforms,
+    get_training_transforms,
+)
+
+__all__ = [
+    "convert_checkpoint",
+    "get_inference_transforms",
+    "get_training_transforms",
+    "denormalize_image",
+]

src/hac/utils/convert_to_safetensors.py renamed to src/hac/common/convert_to_safetensors.py

@@ -0,0 +1,52 @@
+"""Shared metrics for both image and video classification."""
+
+from sklearn.metrics import (
+    accuracy_score,
+    confusion_matrix,
+    precision_recall_fscore_support,
+)
+
+
+def compute_accuracy(predictions, targets):
+    """Compute accuracy.
+
+    Args:
+        predictions: Array of predicted class indices
+        targets: Array of ground truth class indices
+
+    Returns:
+        Accuracy as float
+    """
+    return accuracy_score(targets, predictions)
+
+
+def compute_metrics(predictions, targets, average="macro"):
+    """Compute comprehensive classification metrics.
+
+    Args:
+        predictions: Array of predicted class indices
+        targets: Array of ground truth class indices
+        average: Averaging strategy ('macro', 'weighted', 'micro')
+
+    Returns:
+        Dictionary with accuracy, precision, recall, f1
+    """
+    acc = accuracy_score(targets, predictions)
+    precision, recall, f1, _ = precision_recall_fscore_support(
+        targets, predictions, average=average, zero_division=0
+    )
+
+    return {"accuracy": acc, "precision": precision, "recall": recall, "f1": f1}
+
+
+def compute_confusion_matrix(predictions, targets):
+    """Compute confusion matrix.
+
+    Args:
+        predictions: Array of predicted class indices
+        targets: Array of ground truth class indices
+
+    Returns:
+        Confusion matrix as numpy array
+    """
+    return confusion_matrix(targets, predictions)

src/hac/common/metrics.py

@@ -0,0 +1 @@
+"""Image action recognition module."""