Individual Layers of VGG16
Most Noticeable Eigen-Distortion

Below is presented the largest eigenvector for the Fisher Information matrix (with respect to to this image) of the representations within several layers of VGG16. We present the vectors (top row) and each vector added back to the image at the same amplitude (amplitude =10).

Distortion Vectors

Front	Layer2	Layer3	Layer4	Layer5	Layer6

Images + Vectors

Front	Layer2	Layer3	Layer4	Layer5	Layer6

Individual Layers of VGG16
Least Noticeable Eigen-Distortion

Below is presented the smallest eigenvector for the Fisher Information matrix (with respect to to this image) of the representations within several layers of VGG16. We present the vectors (top row) and each vector added back to the image at the same amplitude (amplitude =30).

Distortion Vectors

Front	Layer2	Layer3	Layer4	Layer5	Layer6

Images + Vectors

Front	Layer2	Layer3	Layer4	Layer5	Layer6

Image Quality Assessment Models
Most Noticeable Eigen-Distortion

Below is presented the largest eigenvector for the Fisher Information matrix (with respect to to this image) of the representations within several optimized Image Quality Assessment Models. We present the vectors (top row) and each vector added back to the image at the same amplitude (amplitude =10).

Distortion Vectors

LN	LG	LGG	On-Off	CNN	VGG-IQA

Images + Vectors

LN	LG	LGG	On-Off	CNN	VGG-IQA

Image Quality Assesment Models
Least Noticeable Eigen-Distortion

Below is presented the smallest eigenvector for the Fisher Information matrix (with respect to to this image) of the representations within several optimized Image Quality Assessment Models. We present the vectors (top row) and each vector added back to the image at the same amplitude (amplitude =30).

Distortion Vectors

LN	LG	LGG	On-Off	CNN	VGG-IQA

Images + Vectors

LN	LG	LGG	On-Off	CNN	VGG-IQA