1. CSSE463: Image Recognition Day 20 Announcements: Announcements: Sunset detector due Weds. 11:59 Sunset detector due Weds. 11:59 Literature reviews due Sunday night 11:59 Literature reviews due Sunday night 11:59 Today: Lab for sunset detector Today: Lab for sunset detector Next week: Next week: Monday: k-means Monday: k-means Tuesday: sunset detector lab Tuesday: sunset detector lab Thursday: template-matching Thursday: template-matching Friday: k-means lab Friday: k-means lab

2. Term Project Teams 1. 1. Photo Mosaic: Ali A, Joe L. Marina K, Daniel N. 2. 2. Where’s Waldo?: Ryohei S, Brandon K, Eric V 3. 3. Stereoscopic Image Depth Mapping: Eric H, Nick K, Laura M, James S 4. 4. Object Tracker: Franklin T, Kyle C, Mike E, Kyle A 5. 5. Detection of Cancer in Brain Scans: Trey C, Nicole R, Katy Y, Maisey T 6. 6. IGVC Vision: Donnie Q, Anders S, Ruffin W, Kurtis Z 7. 7. Paint by Numbers Generator: Ann S, Katie G, Seth T 8. 8. Next Chess Move: Robert W, Andrew M, Eduardo B

3. Term project next steps Read papers! Read papers! Review what others have done. Don’t re-invent the wheel. Review what others have done. Don’t re-invent the wheel. Summarize your papers Summarize your papers Due Due

4. Last words on neural nets

5. Common model of learning machines Statistical Learning (svmtrain) Labeled Training Images Extract Features (color, texture) Test Image Summary Label Classifier (svmfwd) Extract Features (color, texture) Normalize

6. Sunset detector addition Please email me a zip file of 10 images that you think would be tough to classify by Monday. I’ll compile and send out. Please email me a zip file of 10 images that you think would be tough to classify by Monday. I’ll compile and send out. Best if the resolution is 384 x 256 (or slightly bigger, but don’t change the aspect ratio as you rescale) Best if the resolution is 384 x 256 (or slightly bigger, but don’t change the aspect ratio as you rescale)

7. Sunset Process Loop over images Loop over images Extract features Extract features Normalize Normalize Split into train and test Split into train and test Save Save Loop over kernel params Loop over kernel params Train Train Test Test Record accuracy Record accuracy For SVM with param giving best accuracy, For SVM with param giving best accuracy, Generate ROC curve Generate ROC curve Find good images Find good images Do extension Do extension Write as you go! Write as you go!