Test1 -- A : No pretraining + Strictly SUN training images

RankName ApprochAccuracy
1An Nguyen HOG + SIFT + color + linear SVM0.6891
2Tushar Nagarajan Categorial pre-training (from scratch) + AlexNet CNN (from scratch) 0.6112
3Nayan Singhal AlexNet CNN (from scratch) 0.5723
4Vivek Pradhan spatial pyamid features + histogram intersction kernel + SVM0.5488
5Wei-Lin Hsiao PHOW + GMM + linear SVM 0.5184

Test1 -- B: Any pretraining + Strictly SUN training Images

RankName ApprochAccuracy
1Zhenpei Yang Fine tuning on ResNet-1520.9140
2Yiming Pang Fine tuning on ResNet-152 (replace + adding extra 25-way layer) 0.8800
3Hangchen Yu Fine tuning on VGG 0.8750
4Harashal Priyadarshi Fine tuning on GoogleNet (adding extra 25-way layer) 0.8672
5Wenguang Mao Fine tuning on GoogleNet (replace 1000-way to 25-way layer) 0.8530

Test1 -- C: Any pretraining + All provided SUN training Images

RankName ApprochAccuracy
1Yiming Pang Fine tuning on ResNet-152 (fc8 layer) 0.9080
2An Nguyen features from (AlexNet+VGG+ResNet) + linear SVM0.8974
3Zhenpei Yang Fine tuning on ResNet-1520.8950
4Wenguang Mao Fine tuning on GoogleNet (replace 1000-way to 25-way layer) 0.8850
5Hangchen Yu Fine tuning on VGG 0.8830

Test2 -- A: No pretraining + Strictly SUN training images

RankName ApprochAccuracy
1An Nguyen HOG + SIFT + color + linear SVM0.6880
2Tushar Nagarajan Categorial pre-training (from scratch) + AlexNet CNN (from scratch) 0.5792
3Vivek Pradhan spatial pyamid features + histogram intersction kernel + SVM0.5792
4Nayan Singhal AlexNet CNN (from scratch) 0.5536
5Wei-Lin Hsiao PHOW + GMM + linear SVM0.5216

Test2 -- B and C: Any pretraining + Any provided SUN training Images

RankName ApprochAccuracy
1Zhenpei Yang Fine tuning on ResNet-1520.8990
2Hangchen Yu Fine tuning on VGG 0.8910
3An Nguyen features from (AlexNet+VGG+ResNet) + linear SVM0.8864
4Tushar Nagarajan Fine tuning on AlexNet 0.8784
5Yiming Pang Fine tuning on ResNet-152 (fc8 layer) 0.8640