A command line to get the image quality evaluation | with code and Procedure

 

Because after csdn Copy image problem to add: https://cloud.tencent.com/developer/article/1551010

In exchange group, the question is often asked to evaluate the image quality. For example, multiple images captured on surveillance cameras, the selection of an image displayed to the user, or select a picture threw recognition model, or in Internet applications, the image for multiple users to upload, choose one as the cover. General requirements for image clarity, better quality, there is no easy way to achieve image quality assessment of it?

Today recommended a tool with you, from Germany goods comparison service providers idealo open source image quality assessment tool, you need only one command can be achieved.

Open Source Address:

https://github.com/idealo/image-quality-assessment

Installation is very short answer:

To evaluate the quality of an image:

./predict  \
--docker-image nima-cpu \
--base-model-name MobileNet \
--weights-file $(pwd)/models/MobileNet/weights_mobilenet_technical_0.11.hdf5 \
--image-source $(pwd)/src/tests/test_images/42039.jpg

For all the images in a folder in the quality assessment:

./predict  \
--docker-image nima-cpu \
--base-model-name MobileNet \
--weights-file $(pwd)/models/MobileNet/weights_mobilenet_technical_0.11.hdf5 \
--image-source $(pwd)/src/tests/test_images

This tool is very tricky, and its reference is the Google 2017 years of research papers NIMA: Neural Image Assessment "(https://arxiv.org/pdf/1709.05424.pdf), another company itself is its own Internet services this tool is used for selection and recommended users to upload images of the hotel.

The tool has actually aesthetic evaluation (focusing on the image pretty ugly) and technical evaluation (focusing on the good image quality is not good) in two ways.

The official has been given two pre-training model.

Of course, not everyone is doing these two aspects, image quality assessment such as I just said surveillance scene, then you need to own trained.

The authors also provide an easy to use training interface.

After tagging a good sample, configure the environment, training is also only one command:

./train-local \
--config-file $(pwd)/models/MobileNet/config_technical_cpu.json \
--samples-file $(pwd)/data/TID2013/tid_labels_train.json \
--image-dir /path/to/image/dir/local

In short, this is a real project and in research, are noteworthy image quality assessment tool.

Sending again Project Address:

https://github.com/idealo/image-quality-assessment