Details

• Original Images
  - Training negative images: 1218
  - Training positive images: 614
  - Test negative images: 453
  - Test positive images: 288
  
  
• Normalized Images
  - Training positive images: 2416 (images centered on the person with their left-right reflections)
  - Test negative images: 453
  - Test positive images: 1132 (images centered on the person with their left-right reflections)
  
  Note, when you download the images use the centered 64x128 pixels window for original detection task.
  

---

Related papers

• [Dalal05] Navneet Dalal and Bill Triggs. Histograms of Oriented Gradients for Human Detection, France 2005. Download PDF
• Navneet Dalal and Bill Triggs, Phd Thesis. Download PDF

---

Evaluation Protocol

• False Positives in negative images
  
  A fixed set of 12180 windows (10 windows per negative image) are sampled randomly from 1218 negative training photos providing the initial negative training set. For each detector and parameter combination, a preliminary detector is trained and all negative training images are searched exhaustively for false positives.

---

Features and top scores

Method:	[Dalal05]
Score:	89%