A key goal in image restoration, manipulation and generation, is to produce images that are visually appealing to human observers. In recent years, there has been great interest as well as significant progress in perceptually-aware computer vision algorithms. However, many works have observed a fundamental disagreement between this recent leap in performance, as evaluated by human observers, and the objective assessment of these methods by common evaluation metrics (e.g. PSNR, SSIM). This workshop will revolve around two main themes: (i) How to design algorithms which satisfy human observers, and (ii) How to evaluate the perceptual quality of such algorithms.
The workshop will host invited talks by William Freeman (MIT, Google AI), Eli Shechtman (Adobe Research), Tomer Michaeli (Technion), Jiaya Jia (CUHK) and Kavita Bala (Cornell).
Three challenges on perceptual super-resolution, image enhancement on smartphones, and hyperspectral reconstruction will compare and rank perceptual-quality based methods.
Exciting new results of recent and ongoing research projects will be presented in a poster session.
Time |
Event |
|---|---|
8:30 |
Opening |
8:40 |
Invited talk - Jiaya Jia (CUHK) |
9:05 |
Invited talk - Eli Shechtman (Adobe Research) |
9:30 |
Spotlights |
10:00 |
Poster session and coffee break |
10:45 |
Invited talk - Kavita Bala (Cornell) |
11:10 |
Invited talk - Tomer Michaeli (Technion) |
11:35 |
The PIRM challenges on: |
12:05 |
Invited talk - William Freeman (MIT, Google AI) |
12:30 |
Closing |
The PIRM challenges will promote perceptual image restoration. State-of-the-art methods in terms of perceptual quality (e.g. SRGAN) are rated poorly by "simple" distortion measures such as PSNR and SSIM. Therefore, in contrast to previous challenges, the evaluation and ranking will be done in a perceptual-quality aware manner based on [Blau and Michaeli, CVPR'18]. This unified approach quantifies the accuracy and perceptual quality of algorithms jointly, and will enable perceptual-driven methods to compete alongside algorithms that target PSNR maximization.
The challenge results will be announced at the workshop. Challenge participants may submit papers for the ECCV workshop proceedings. Papers will be accepted based on: (i) academic quality, and (ii) challenge ranking.
Date |
Event |
|---|---|
May 1st |
Validation data released |
July 18th |
Test data released |
July 25th |
Final results submission deadline |
July 29th |
Poster submission deadline |
August 1st |
Challenge results released to participants |
August 22nd |
Paper submission deadline |
September 5th |
Notification of accepted papers |
September 14th |
PIRM 2018 Workshop |
September 30th |
Camera-ready deadline |
ETH Zurich, Switzerland
Data61-CSIRO
