investigate very low accuracy class prediction images in MSRC

RockStarCoders / alienMarkovNetworks

Using MRFs and CRFs for computer vision problems.

21 stars 9 forks source link

investigate very low accuracy class prediction images in MSRC #20

Closed jsherrah closed 11 years ago

jsherrah commented 11 years ago

0% would be a good start :)

amb-enthusiast commented 11 years ago

Used evalPredictions.py on our reference prediction images and logged all images with prediction accuracy of 0% (see zeroAccuracyList.txt). Classifier struggles with bird, chair, flower and dog classification - likely due the reliance on colour-derived features.