En cours

Project for lakshyakejriwal

project is about creating an image repository and training 7 different photo models (different data sets) for waifu2x including scale 2x, noise 1 and noise 2 and combined scale and noise model. All models are photographic models. 

The models that need to be trained will be of hierarchical nature. We might train submodels in the future such as football model as submodel of sports model etc.:

Photo (trained on all images from below submodels)

  Food and Drink 


  People portraits



  Architecture and Design 

First we need to create data repostitory for images. My idea is to download several image sources and create subfolders with the above mentioned categories. 

Shutterstock photos are already classified in these categories. I have downloaded parts 1 and 2:

[url removed, login to view]

Open image download. Not sure how the images are categorized.

[url removed, login to view] 

Perhaps you have an idea about anothr good dataset. 

I [url removed, login to view] each model should be teained roughly on 5000-10000 images.

All models need to be tested on our standard image datasets and domain specific datasets that we create.

We need a scrpt that test specified model and outputs the testimages. I assume that you have done it but i wouldnlike to agree on folder structure e.g ../test/TESTSET (such as S5)/MODELNAME(need to decide on naming convention)

The benchmarking script should be able to be a nice script that takes speicified  folders of test images described above from commandline and output all relevant benchmark stats such as PSNR and time to completion, testing machine [url removed, login to view] a nice table csv format. I assume you have written it partly for this project.

We also need to agree on a naming convention for each new model created. Any suggestions?

Retarding image categorization into test folders please let me know if you need some help. We can get a data input person helping you on this.

Before starting we need to make sure that our own trained model and default model on waifu give same / similar test results.

Please let me know what you think. Any suggestions are welcome. Also please let me know a project fee for this project. Project 3 will be fine tuning these models or image classification into the groups mentioned above.

Compétences : Analyses statistiques

en voir plus : s&e partners, s.e.m. technologies, http www cracksinn com p s e logo design st, d e s i g n l o g o, d e s i g n - l o g o ? ? ? ? ? ? ? ? ?, project data structure java, struts ejb hibernate project structure, project portal structure, data structure project java railway system country, data structure java project, structure website project aspnet, phpmysql structure big project, project structure chart, programming project data structure, data structure project

Concernant l'employeur :
( 66 commentaires ) London, United Kingdom

Nº du projet : #12643176

Décerné à:


Hired by the Employer

%selectedBids___i_sum_sub_7% %project_currencyDetails_sign_sub_8% USD en 7 jours
(2 Commentaires)