MultiGrain: a unified image embedding for classes and instances

14 Feb 2019Maxim BermanHervé JégouAndrea VedaldiIasonas KokkinosMatthijs Douze

MultiGrain is a network architecture producing compact vector representations that are suited both for image classification and particular object retrieval. It builds on a standard classification trunk... (read more)

PDF Abstract

Evaluation Results from the Paper


TASK DATASET MODEL METRIC NAME METRIC VALUE GLOBAL RANK COMPARE
Image Classification ImageNet MultiGrain PNASNet (500px) Top 1 Accuracy 83.6% # 22
Image Classification ImageNet MultiGrain PNASNet (500px) Top 5 Accuracy 96.7% # 16
Image Classification ImageNet MultiGrain PNASNet (500px) Number of params 86M # 1
Image Classification ImageNet MultiGrain NASNet-A-Mobile (350px) Top 1 Accuracy 75.1% # 100
Image Classification ImageNet MultiGrain NASNet-A-Mobile (350px) Top 5 Accuracy 92.5% # 77
Image Classification ImageNet MultiGrain R50-AA-224 Top 1 Accuracy 78.2% # 76
Image Classification ImageNet MultiGrain R50-AA-224 Top 5 Accuracy 93.9% # 60
Image Classification ImageNet MultiGrain SENet154 (400px) Top 1 Accuracy 83.0% # 25
Image Classification ImageNet MultiGrain SENet154 (400px) Top 5 Accuracy 96.5% # 18
Image Classification ImageNet MultiGrain SENet154 (400px) Number of params 86M # 1
Image Classification ImageNet MultiGrain R50-AA-500 Top 1 Accuracy 79.4% # 54
Image Classification ImageNet MultiGrain R50-AA-500 Top 5 Accuracy 94.8% # 40
Image Retrieval INRIA Holidays MultiGrain R50 @ 500 Mean mAP 91.8% # 2
Image Retrieval INRIA Holidays MultiGrain R50 @ 800 Mean mAP 92.5% # 1