Skip to content

README.md

Latest commit

 

History

History
119 lines (86 loc) · 3.23 KB

README.md

File metadata and controls

119 lines (86 loc) · 3.23 KB
Raw

Ad Tracking Pipeline

Automatic detection and tracking of advertisements in screen recordings using SAM2.

Requirements

pip install -r requirements.txt

Pipeline

1. Annotate Ads

Manually mark all advertisements in your video:

python code/annotate_video.py --video "data/input/your_video.mp4"

Tips:

  • Set window to fullscreen for better visualization
  • Mark the first frame where each ad appears (will be tracked forward based on seconds indicated in Step 2 config)
  • Follow commands shown in terminal
  • After confirming a bbox, a popup will ask for a name for the ad

Output: data/output/{video_name}/ads_annotations.txt

2. Process All Ads

Automatically segment all annotated ads with SAM2:

# Single video
python code/inference_on_queue.py --video "data/input/your_video.mp4"

# Batch mode (all videos with annotations)
python code/inference_on_queue.py --all

Configuration (in detect_one_ad.py):

SAM_MODEL_PATH = "sam2.1_s.pt"  # small model, change 's' to 'b' for more precision
TIME_WINDOW_AFTER = 15.0  # seconds to track forward

Tips:

  • If interrupted, the script can resume from where it stopped
  • Answer 'y' to skip already processed ads, 'n' to reprocess all

The script will:

  • Find existing processed ads
  • Process ads with SAM2
  • Save individual JSON and update master file after each ad

Output:

  • data/output/{video_name}/detections/ad_*_detections.json (individual)
  • data/output/{video_name}/all_detections.txt (master registry)

3. Visualize Results

Generate annotated video with all detections:

python code/visualize_final_result.py --video "data/input/your_video.mp4"

Output: data/output/{video_name}/{video_name}_annotated.mp4

4. Modify Detections (Optional)

Interactively review and edit detections:

python code/modify_ads.py --video "data/input/your_video.mp4"

⚠️ Backup all_detections.txt and detections/ folder before using!

Features:

  • Navigate between ads
  • Modify and delete boxes in each frame
  • Edit names and delete entire ads
  • Changes only saved when pressing s or confirming on exit

Output: Modified files in data/output/{video_name}/detections/ and all_detections.txt

Directory Structure

ad_tracking/
├── data/
│   ├── input/
│   │   └── your_video.mp4          # Input video
│   └── output/
│       └── {video_name}/
│           ├── ads_annotations.txt       # Manual annotations
│           ├── detections/
│           │   ├── ad_1_detections.json  # Per-ad detections
│           │   ├── ad_2_detections.json
│           │   └── ...
│           ├── all_detections.txt        # Master registry
│           └── {video_name}_annotated.mp4  # Final video
├── code/
│   ├── annotate_video.py           # Step 1
│   ├── detect_one_ad.py            # Step 2 (used by inference_on_queue)
│   ├── inference_on_queue.py       # Step 2
│   ├── modify_ads.py               # Step 4 (optional)
│   └── visualize_final_result.py   # Step 3
└── sam2.1_s.pt                     # SAM2 model (small, use sam2.1_b.pt for base)