Skip to main content

Table 11 MAP evaluations for the large-scale experiments with MIR Flickr image distractors

From: Localizing global descriptors for content-based image retrieval

  UKBench UCID
     +     +  
   Cb Dataset Distract. Loss Cb Dataset Distract. Loss
SIFT CEDD 512 0.8136 0.8072 0.8 % 128 0.6813 0.6218 8.7 %
  SCD 512 0.8764 0.8208 6.3 % 512 0.7145 0.6523 8.7 %
SURF CEDD 512 0.8964 0.8712 2.8 % 2048 0.7811 0.6951 11.0 %
  SCD 512 0.9145 0.8466 7.4 % 2048 0.7718 0.6932 10.2 %
Rnd CEDD 2048 0.9183 0.9009 1.9 % 2048 0.7890 0.7001 11.3 %
  SCD 2048 0.9268 0.8869 4.3 % 2048 0.7794 0.6981 10.4 %
Gauss CEDD 2048 0.9245 0.8956 3.1 % 2048 0.7955 0.7028 11.7 %
Rnd SCD 2048 0.9254 0.8884 4.0 % 2048 0.7876 0.7066 10.3 %
   Holidays ZuBuD
     +     +  
   Cb Dataset Distract. Loss Cb Dataset Distract. Loss
SIFT CEDD 512 0.7441 0.7082 4.8 % 2048 0.6854 0.6422 6.3 %
  SCD 512 0.7506 0.7064 5.9 % 512 0.5451 0.5173 5.1 %
SURF CEDD 2048 0.7763 0.7528 3.0 % 2048 0.8340 0.7567 9.3 %
  SCD 512 0.7531 0.7237 3.9 % 2048 0.7453 0.6900 7.4 %
Rnd CEDD 2048 0.8077 0.7633 5.5 % 2048 0.8338 0.7744 7.1 %
  SCD 2048 0.8042 0.7462 7.2 % 2048 0.8287 0.7626 8.0 %
Gauss CEDD 2048 0.8172 0.7545 7.7 % 2048 0.8287 0.7731 6.7 %
Rnd SCD 2048 0.7968 0.7277 8.7 % 2048 0.8117 0.7571 6.7 %