Image & Video Anonymization Tool

Interactive object segmentation and anonymization using SAM, KLT tracking, and optical flow.

---

Quick Start

1. Setup

Install dependencies:

pip install -r requirements.txt

Ensure the SAM checkpoint exists:

model/checkpoints/sam_vit_b_01ec64.pth

2. Prepare Input

• Images: Place files in the IMG/ folder
• Videos: Place files in the VID/ folder

Sample files are already included in both folders for immediate testing.

3. Configure Input File

Open main.py and specify your input:

• For Image mode (line ~1927):

  selected_image = "YOUR_IMAGE_FROM_IMG_FOLDER.jpeg"

• For Video mode (line ~2072):

  selected_video = "YOUR_VIDEO_FROM_VID_FOLDER.mp4"

4. Run

python main.py

---

Interactive Workflow

Step 1: Select Processing Mode

Choose between:

1. Image – Process a single image
2. Video – Process video with frame-by-frame tracking

Step 2: Select ROI (Region of Interest)

• A window displays your image (or video's first frame)
• Drag a rectangle around the object to anonymize
• Press ENTER or ESC to confirm

Step 3: Choose Mask Initialization Method

Select how the mask is generated:

1. SAM (Segment Anything Model)
   AI-powered segmentation—automatic and accurate

2. Shi–Tomasi Corner Detection + Convex Hull
   Classical feature-based approach—lightweight

Step 4 (Video Only): Choose Tracking Method

Select how the mask is propagated across frames:

1. Dense Optical Flow (Farnebäck)
   Pixel-wise motion warping—handles deformations

2. KLT (Kanade–Lucas–Tomasi)
   Feature point tracking with global transform—fast and stable

---

Outputs

• Images: Saved to IMG/ with filename pattern:
  {original_name}_anonymized_{timestamp}.png

• Videos: Saved to VID/ with filename pattern:
  {original_name}_blur_{tracker}_{timestamp}.mp4