singlebehaviorlab 2.0.0__py3-none-any.whl

singlebehaviorlab/backend/video_utils.py

@@ -0,0 +1,139 @@
+import logging
+import cv2
+import os
+from typing import Optional
+
+logger = logging.getLogger(__name__)
+
+
+def extract_clips(
+    video_path: str,
+    output_dir: str,
+    target_fps: int = 16,
+    clip_length_frames: int = 16,
+    step_frames: int = 16,
+    progress_callback: Optional[callable] = None,
+    stop_callback: Optional[callable] = None,
+) -> tuple[int, str]:
+    """Subsample video to target_fps and cut into non-overlapping clips (use step_frames < clip_length_frames for overlap)."""
+    os.makedirs(output_dir, exist_ok=True)
+    
+    cap = cv2.VideoCapture(video_path)
+    if not cap.isOpened():
+        raise ValueError(f"Could not open video: {video_path}")
+
+    try:
+        orig_fps = cap.get(cv2.CAP_PROP_FPS)
+        if orig_fps <= 0:
+            orig_fps = 30.0
+
+        frame_interval = max(1, int(round(orig_fps / target_fps)))
+        total_frames = int(cap.get(cv2.CAP_PROP_FRAME_COUNT))
+
+        total_frames_after_subsampling = total_frames // frame_interval
+        if step_frames >= clip_length_frames:
+            total_clips = total_frames_after_subsampling // clip_length_frames
+        else:
+            total_clips = max(0, (total_frames_after_subsampling - clip_length_frames) // step_frames + 1) if total_frames_after_subsampling >= clip_length_frames else 0
+
+        frame_idx = 0
+        clip_idx = 0
+        frames_buffer = []
+        skip_remaining = 0
+
+        while True:
+            if stop_callback and stop_callback():
+                break
+            ret, frame = cap.read()
+            if not ret:
+                break
+
+            if frame_idx % frame_interval == 0:
+                if skip_remaining > 0:
+                    skip_remaining -= 1
+                else:
+                    frames_buffer.append(frame)
+
+                if len(frames_buffer) == clip_length_frames:
+                    clip_path = os.path.join(output_dir, f"clip_{clip_idx:06d}.mp4")
+                    save_clip(frames_buffer, clip_path, target_fps)
+                    clip_idx += 1
+
+                    if progress_callback:
+                        progress_callback(clip_idx, total_clips)
+
+                    if step_frames < clip_length_frames:
+                        frames_buffer = frames_buffer[clip_length_frames - step_frames:]
+                    else:
+                        frames_buffer = []
+                        skip_remaining = max(0, step_frames - clip_length_frames)
+
+            frame_idx += 1
+
+        return clip_idx, output_dir
+    finally:
+        cap.release()
+
+
+def save_clip(frames: list, output_path: str, fps: float):
+    """Save a list of frames as a standard MP4 clip (mp4v codec)."""
+    if not frames:
+        return
+    
+    h, w, c = frames[0].shape
+    fourcc = cv2.VideoWriter_fourcc(*'mp4v')
+    out = cv2.VideoWriter(output_path, fourcc, fps, (w, h))
+    
+    for frame in frames:
+        out.write(frame)
+    
+    out.release()
+
+
+def load_clip_frames(clip_path: str, target_size: Optional[tuple[int, int]] = None) -> list:
+    """Load frames from a video clip.
+    
+    Args:
+        clip_path: Path to video clip
+        target_size: Optional (width, height) to resize frames
+    
+    Returns:
+        List of frames as numpy arrays (BGR format)
+    """
+    cap = cv2.VideoCapture(clip_path)
+    frames = []
+
+    try:
+        while True:
+            ret, frame = cap.read()
+            if not ret:
+                break
+
+            if target_size:
+                frame = cv2.resize(frame, target_size, interpolation=cv2.INTER_AREA)
+
+            frames.append(frame)
+
+        return frames
+    finally:
+        cap.release()
+
+
+def get_video_info(video_path: str) -> dict:
+    """Get video metadata."""
+    cap = cv2.VideoCapture(video_path)
+    if not cap.isOpened():
+        return {}
+
+    try:
+        info = {
+            'fps': cap.get(cv2.CAP_PROP_FPS),
+            'frame_count': int(cap.get(cv2.CAP_PROP_FRAME_COUNT)),
+            'width': int(cap.get(cv2.CAP_PROP_FRAME_WIDTH)),
+            'height': int(cap.get(cv2.CAP_PROP_FRAME_HEIGHT)),
+        }
+
+        return info
+    finally:
+        cap.release()
+

singlebehaviorlab/data/config/config.yaml

@@ -0,0 +1,85 @@
+# SingleBehaviorLab Configuration
+# Paths below are resolved relative to the app directory on first launch.
+# They will be updated automatically when you create or load an experiment.
+
+# --- Paths (leave blank; auto-resolved at startup) ---
+data_dir:
+raw_videos_dir:
+clips_dir:
+annotations_dir:
+models_dir:
+backbone_dir:
+annotation_file:
+experiments_dir:
+experiment_name:
+experiment_path:
+config_path:
+
+# --- Video extraction defaults ---
+default_target_fps: 12
+default_clip_length: 8
+default_step_frames: 8
+
+# --- Training defaults ---
+default_batch_size: 16
+default_epochs: 60
+default_learning_rate: 0.0005
+default_weight_decay: 0.001
+
+# --- Model ---
+videoprism_model_name: videoprism_public_v1_base
+backbone_model: videoprism_public_v1_base
+resolution: 288
+model_input_size:
+  - 288
+  - 288
+
+# --- Training options ---
+default_use_scheduler: true
+default_use_ema: true
+default_map_num_queries: 1
+default_use_class_map: false
+default_use_cosine: false
+default_use_label_smoothing: false
+default_use_supcon_loss: true
+default_supcon_temp: 0.07
+default_supcon_weight: 0.5
+default_use_focal_loss: true
+default_focal_gamma: 2.0
+
+# --- Last training session (auto-updated by app) ---
+last_training:
+  parameters:
+    batch_size: 16
+    epochs: 60
+    lr: 0.0005
+    use_scheduler: true
+    weight_decay: 0.001
+    clip_length: 8
+    val_split: 0.2
+    use_class_weights: false
+    use_label_smoothing: false
+    use_supcon_loss: true
+    supcon_temp: 0.07
+    supcon_weight: 0.5
+    use_triplet_loss: false
+    triplet_margin: 0.2
+    triplet_weight: 0.5
+    use_focal_loss: true
+    focal_gamma: 2.0
+    use_weighted_sampler: true
+    use_augmentation: true
+    limit_classes: true
+    limit_per_class: false
+  selected_classes: []
+  per_class_limits: {}
+  head:
+    dropout: 0.25
+    map_head_kwargs:
+      num_heads: 4
+      mlp_ratio: 4.0
+      num_queries: 1
+      use_class_map: false
+      use_cosine: false
+  pretrained_path:
+  output_path:

singlebehaviorlab/data/training_profiles.json

@@ -0,0 +1,334 @@
+{
+  "LowInputData": {
+    "batch_size": 8,
+    "epochs": 60,
+    "lr": 5e-05,
+    "classification_lr": 5e-05,
+    "use_scheduler": true,
+    "use_ema": true,
+    "weight_decay": 0.01,
+    "head_kwargs": {
+      "num_heads": 1
+    },
+    "dropout": 0.5,
+    "clip_length": 8,
+    "use_all_for_training": true,
+    "val_split": 0.15,
+    "auto_tune_before_final": false,
+    "auto_tune_runs": 8,
+    "auto_tune_epochs": 12,
+    "use_class_weights": false,
+    "use_focal_loss": false,
+    "focal_gamma": 2.0,
+    "use_supcon_loss": false,
+    "supcon_weight": 0.8,
+    "supcon_temperature": 0.1,
+    "use_frame_loss": true,
+    "use_temporal_decoder": true,
+    "frame_head_temporal_layers": 1,
+    "temporal_pool_frames": 1,
+    "proj_dim": 64,
+    "use_frame_bout_balance": true,
+    "frame_bout_balance_power": 0.5,
+    "boundary_loss_weight": 0.3,
+    "boundary_tolerance": 2,
+    "smoothness_loss_weight": 0.05,
+    "use_ovr": true,
+    "ovr_background_as_negative": false,
+    "ovr_background_class_names": [],
+    "ovr_label_smoothing": 0.05,
+    "use_asl": false,
+    "asl_gamma_neg": 3.0,
+    "asl_gamma_pos": 0.0,
+    "asl_clip": 0.05,
+    "use_hard_pair_mining": false,
+    "hard_pairs": [],
+    "hard_pair_loss_weight": 0.2,
+    "hard_pair_margin": 0.5,
+    "hard_pair_confusion_boost": 1.5,
+    "use_confusion_sampler": true,
+    "confusion_sampler_temperature": 2.0,
+    "confusion_sampler_warmup_pct": 0.2,
+    "use_weighted_sampler": true,
+    "use_augmentation": true,
+    "virtual_expansion": 15,
+    "stitch_augmentation_prob": 0.0,
+    "emb_aug_versions": 5,
+    "multi_scale": false,
+    "augmentation_options": {
+      "use_horizontal_flip": true,
+      "use_vertical_flip": true,
+      "use_color_jitter": true,
+      "use_gaussian_blur": true,
+      "use_random_noise": true,
+      "use_small_rotation": true,
+      "use_speed_perturb": false,
+      "use_random_shapes": true,
+      "use_grayscale": false,
+      "use_lighting_robustness": true
+    },
+    "limit_classes": false,
+    "selected_classes": [],
+    "limit_per_class": false,
+    "per_class_limits": {},
+    "per_class_val_limits": {},
+    "use_embedding_diversity": false,
+    "backbone_model": "videoprism_public_v1_base",
+    "class_names": [
+      "walk",
+      "nowalk"
+    ],
+    "pretrained_path": null,
+    "f1_exclude_classes": [],
+    "ovr_pos_weight_f1_excluded": 1.5
+  },
+  "MoreInputData": {
+    "batch_size": 24,
+    "epochs": 60,
+    "lr": 5e-05,
+    "classification_lr": 5e-05,
+    "use_scheduler": true,
+    "use_ema": true,
+    "weight_decay": 0.01,
+    "head_kwargs": {
+      "num_heads": 8
+    },
+    "dropout": 0.6,
+    "clip_length": 8,
+    "use_all_for_training": false,
+    "val_split": 0.15,
+    "auto_tune_before_final": false,
+    "auto_tune_runs": 8,
+    "auto_tune_epochs": 12,
+    "use_class_weights": false,
+    "use_focal_loss": false,
+    "focal_gamma": 2.0,
+    "use_supcon_loss": true,
+    "supcon_weight": 0.8,
+    "supcon_temperature": 0.1,
+    "use_frame_loss": true,
+    "use_temporal_decoder": true,
+    "frame_head_temporal_layers": 6,
+    "temporal_pool_frames": 1,
+    "proj_dim": 320,
+    "use_frame_bout_balance": true,
+    "frame_bout_balance_power": 0.5,
+    "boundary_loss_weight": 0.3,
+    "boundary_tolerance": 2,
+    "smoothness_loss_weight": 0.05,
+    "use_ovr": true,
+    "ovr_background_as_negative": false,
+    "ovr_background_class_names": [],
+    "ovr_label_smoothing": 0.05,
+    "use_asl": false,
+    "asl_gamma_neg": 3.0,
+    "asl_gamma_pos": 0.0,
+    "asl_clip": 0.05,
+    "use_hard_pair_mining": false,
+    "hard_pairs": [],
+    "hard_pair_loss_weight": 0.2,
+    "hard_pair_margin": 0.5,
+    "hard_pair_confusion_boost": 1.5,
+    "use_confusion_sampler": true,
+    "confusion_sampler_temperature": 2.0,
+    "confusion_sampler_warmup_pct": 0.2,
+    "use_weighted_sampler": true,
+    "use_augmentation": true,
+    "virtual_expansion": 10,
+    "stitch_augmentation_prob": 0.0,
+    "emb_aug_versions": 5,
+    "multi_scale": false,
+    "augmentation_options": {
+      "use_horizontal_flip": true,
+      "use_vertical_flip": true,
+      "use_color_jitter": true,
+      "use_gaussian_blur": true,
+      "use_random_noise": true,
+      "use_small_rotation": true,
+      "use_speed_perturb": false,
+      "use_random_shapes": true,
+      "use_grayscale": false,
+      "use_lighting_robustness": true
+    },
+    "limit_classes": false,
+    "selected_classes": [],
+    "limit_per_class": false,
+    "per_class_limits": {},
+    "per_class_val_limits": {},
+    "use_embedding_diversity": false,
+    "backbone_model": "videoprism_public_v1_base",
+    "class_names": [
+      "walk",
+      "nowalk"
+    ],
+    "pretrained_path": null,
+    "f1_exclude_classes": [],
+    "ovr_pos_weight_f1_excluded": 1.5
+  },
+  "LocalizationLowData": {
+    "batch_size": 8,
+    "epochs": 60,
+    "lr": 5e-05,
+    "classification_lr": 5e-05,
+    "use_scheduler": true,
+    "use_ema": true,
+    "weight_decay": 0.01,
+    "head_kwargs": {
+      "num_heads": 1
+    },
+    "dropout": 0.5,
+    "clip_length": 8,
+    "use_all_for_training": true,
+    "val_split": 0.15,
+    "auto_tune_before_final": false,
+    "auto_tune_runs": 8,
+    "auto_tune_epochs": 12,
+    "use_class_weights": false,
+    "use_focal_loss": false,
+    "focal_gamma": 2.0,
+    "use_supcon_loss": false,
+    "supcon_weight": 0.8,
+    "supcon_temperature": 0.1,
+    "use_frame_loss": true,
+    "use_temporal_decoder": true,
+    "frame_head_temporal_layers": 1,
+    "temporal_pool_frames": 1,
+    "proj_dim": 64,
+    "use_frame_bout_balance": true,
+    "frame_bout_balance_power": 0.5,
+    "boundary_loss_weight": 0.3,
+    "boundary_tolerance": 2,
+    "smoothness_loss_weight": 0.05,
+    "use_ovr": true,
+    "ovr_background_as_negative": false,
+    "ovr_background_class_names": [],
+    "ovr_label_smoothing": 0.05,
+    "use_asl": false,
+    "asl_gamma_neg": 3.0,
+    "asl_gamma_pos": 0.0,
+    "asl_clip": 0.05,
+    "use_hard_pair_mining": false,
+    "hard_pairs": [],
+    "hard_pair_loss_weight": 0.2,
+    "hard_pair_margin": 0.5,
+    "hard_pair_confusion_boost": 1.5,
+    "use_confusion_sampler": true,
+    "confusion_sampler_temperature": 2.0,
+    "confusion_sampler_warmup_pct": 0.2,
+    "use_weighted_sampler": true,
+    "use_augmentation": true,
+    "virtual_expansion": 15,
+    "stitch_augmentation_prob": 0.0,
+    "emb_aug_versions": 5,
+    "multi_scale": false,
+    "augmentation_options": {
+      "use_horizontal_flip": true,
+      "use_vertical_flip": true,
+      "use_color_jitter": true,
+      "use_gaussian_blur": true,
+      "use_random_noise": true,
+      "use_small_rotation": true,
+      "use_speed_perturb": false,
+      "use_random_shapes": true,
+      "use_grayscale": false,
+      "use_lighting_robustness": true
+    },
+    "limit_classes": false,
+    "selected_classes": [],
+    "limit_per_class": false,
+    "per_class_limits": {},
+    "per_class_val_limits": {},
+    "use_embedding_diversity": false,
+    "backbone_model": "videoprism_public_v1_base",
+    "class_names": [
+      "walk",
+      "nowalk"
+    ],
+    "pretrained_path": null,
+    "f1_exclude_classes": [],
+    "ovr_pos_weight_f1_excluded": 1.5
+  },
+  "LocalizationMoreData": {
+    "batch_size": 24,
+    "epochs": 60,
+    "lr": 5e-05,
+    "classification_lr": 5e-05,
+    "use_scheduler": true,
+    "use_ema": true,
+    "weight_decay": 0.01,
+    "head_kwargs": {
+      "num_heads": 8
+    },
+    "dropout": 0.6,
+    "clip_length": 8,
+    "use_all_for_training": false,
+    "val_split": 0.15,
+    "auto_tune_before_final": false,
+    "auto_tune_runs": 8,
+    "auto_tune_epochs": 12,
+    "use_class_weights": false,
+    "use_focal_loss": false,
+    "focal_gamma": 2.0,
+    "use_supcon_loss": true,
+    "supcon_weight": 0.8,
+    "supcon_temperature": 0.1,
+    "use_frame_loss": true,
+    "use_temporal_decoder": true,
+    "frame_head_temporal_layers": 6,
+    "temporal_pool_frames": 1,
+    "proj_dim": 320,
+    "use_frame_bout_balance": true,
+    "frame_bout_balance_power": 0.5,
+    "boundary_loss_weight": 0.3,
+    "boundary_tolerance": 2,
+    "smoothness_loss_weight": 0.05,
+    "use_ovr": true,
+    "ovr_background_as_negative": false,
+    "ovr_background_class_names": [],
+    "ovr_label_smoothing": 0.05,
+    "use_asl": false,
+    "asl_gamma_neg": 3.0,
+    "asl_gamma_pos": 0.0,
+    "asl_clip": 0.05,
+    "use_hard_pair_mining": false,
+    "hard_pairs": [],
+    "hard_pair_loss_weight": 0.2,
+    "hard_pair_margin": 0.5,
+    "hard_pair_confusion_boost": 1.5,
+    "use_confusion_sampler": true,
+    "confusion_sampler_temperature": 2.0,
+    "confusion_sampler_warmup_pct": 0.2,
+    "use_weighted_sampler": true,
+    "use_augmentation": true,
+    "virtual_expansion": 10,
+    "stitch_augmentation_prob": 0.0,
+    "emb_aug_versions": 5,
+    "multi_scale": false,
+    "augmentation_options": {
+      "use_horizontal_flip": true,
+      "use_vertical_flip": true,
+      "use_color_jitter": true,
+      "use_gaussian_blur": true,
+      "use_random_noise": true,
+      "use_small_rotation": true,
+      "use_speed_perturb": false,
+      "use_random_shapes": true,
+      "use_grayscale": false,
+      "use_lighting_robustness": true
+    },
+    "limit_classes": false,
+    "selected_classes": [],
+    "limit_per_class": false,
+    "per_class_limits": {},
+    "per_class_val_limits": {},
+    "use_embedding_diversity": false,
+    "backbone_model": "videoprism_public_v1_base",
+    "class_names": [
+      "walk",
+      "nowalk"
+    ],
+    "pretrained_path": null,
+    "f1_exclude_classes": [],
+    "ovr_pos_weight_f1_excluded": 1.5
+  }
+}

singlebehaviorlab/gui/__init__.py

@@ -0,0 +1,4 @@
+from .main_window import MainWindow
+
+__all__ = ['MainWindow']
+