Visual Saliency: From Pixel-Level to Object-Level Analysis by Jianming Zhang & Filip Malmberg & Stan Sclaroff

Author:Jianming Zhang & Filip Malmberg & Stan Sclaroff
Language: eng
Format: epub
ISBN: 9783030048310
Publisher: Springer International Publishing

CNN_FT: The CNN model fine-tuned on the real image data only.

CNN_Syn: The CNN model fine-tuned on the synthetic images only. This baseline reflects how close the synthetic images are to the real data.

CNN_wo_FT: The features of the pre-trained GoogleNet without fine-tuning. For this baseline, we fix the parameters of all the hidden layers during fine-tuning. In other words, only the output layer is fine-tuned.

Furthermore, we benchmark several commonly used image feature representations for baseline comparison. For each feature representation, we train a one-vs-all multi-class linear SVM classifier on the training set. The hyper-parameters of the SVM are determined via fivefold cross-validation. GIST. The GIST descriptor [177] is computed based on 32 Gabor-like filters with varying scales and orientations. We use the implementation by [177] to extract a 512-D GIST feature, which is a concatenation of averaged filter responses over a 4 × 4 grid.

Download

Copyright Disclaimer:
This site does not store any files on its server. We only index and link to content provided by other sites. Please contact the content providers to delete copyright contents if any and email us, we'll remove relevant links or contents immediately.