Trainable OCR Winner

 

The Winning Entry

The winning entry was ntu.csie.DDK10, supplied by Chih-Chung Chang of the SVM group, department of Computer Science and Information Engineering, National Taiwan University. 

On PO Digits, after 100 experimental evaluations, the top two entries were:

rootClass nTrials Code size Trained size Train time Rec time Save time Load time Train Accuracy Test Accuracy
ntu.csie.RCO10310310 100 12921.0  (0.) 11975.8  (86.59) 28.69  (4.55) 31.12  (3.12) 112.0  (9.84) 5279.2  (431.57) 100.0  (0.) 92.69  (1.73)
ntu.csie.DDK10 95 7939.0  (0.) 15263.4  (139.75) 5.26  (1.13) 17.15  (1.97) 177.05  (39.89) 163.15  (65.02) 100.0  (0.) 92.30  (1.71)

RCO... was not significantly more accurate (at 95% confidence level), so the award goes to DDK10 with the faster recognition time.

Algoval salutes you!