莊 永 裕 國立台灣大學 資訊工程學系 通訊與多媒體實驗室 2010年專題研究說明 莊 永 裕 國立台灣大學 資訊工程學系 通訊與多媒體實驗室 Thanks for the introduction. I was really busy the last few weeks and I just didn’t have time to prepare this talk well. But I do have some great pictures of my trip to Japan!
Research topics Rendering VFX and computational photography Content analysis for Surveillance and Multimedia
Rendering 3D model rendering
Pixar
Renderman
Time-consuming 2 hours, 1995 15 hours, 2005 During Ratatouille, the renderfarm consisted of about 850 machines with nearly 3200 processors between them. When rendering the final Ratatouille film frames on a 2.66 GHz processor, each frame took an average of six hours. It took about 1532 CPU-years to render Ratatouille, including the lower-resolution renders done at various points in the pipeline and working iterations. That means that if we only had one CPU in the renderfarm, Ratatouille wouldn’t have been released until the year 3539 (1532 years). To store the images generated while making the movie, we used 12 terabytes of disk space. 2 hours, 1995 15 hours, 2005
Pixar’s render farm 3,200 processors for Ratatouille (2007) 1,600 square feet 335kW IT load During Ratatouille, the renderfarm consisted of about 850 machines with nearly 3200 processors between them. When rendering the final Ratatouille film frames on a 2.66 GHz processor, each frame took an average of six hours. It took about 1532 CPU-years to render Ratatouille, including the lower-resolution renders done at various points in the pipeline and working iterations. That means that if we only had one CPU in the renderfarm, Ratatouille wouldn’t have been released until the year 3539 (1532 years). To store the images generated while making the movie, we used 12 terabytes of disk space. It will take 1,500 years to render if using only 1 CPU.
Expensive RenderMan Pro Server costs $3,500 USD (~0.1M NTD) per license. It costs roughly 1000*0.1=100M per year for license fee only. 4 threads per license, 3500USD~0.1M NTD 5000 processors for ILM’s
Old architecture
Digimax
Rendering software Develop an in-house renderer Utilize GPUs Include global illumination
VFX and computational photography
HDR
HDR
Hand-held cameras
HDR (Hand-held cameras)
Hand-held camera HDR
Hand-held camera HDR
Video stabilization original video stabilized video
3D era
3D displays
Conversion
Conversion
3D cameras
3D media editing Stabilization Retargeting Slideshow …
Content analysis
Face detection
Gender classification
People counting
Choose a topic Solve a problem of your own interest Cooperate with senior students on his projects Be assigned a topic or a paper to implement A PhD student will be assigned as a mentor for each student
Meetings Biweekly senior project meeting Weekly group meeting Team meeting (optional)
Fin. Thanks.
Pedestrian detection
Event detection
Short shutter (1/512 s)
Content analysis Sports Entertainment Weather Court Office Meeting Studio Outdoor Building Desert Vegetation Mountain Road Sky Snow Urban Waterscape_ Waterfront Crowd Face Person Government- Leader Corporate- Police_ Security Military Prisoner Animal Computer_ TV-screen Flag-US Airplane Car Bus Truck Boat_Ship Walking_ Running People-Marching Explosion_ Fire Natural- Disaster Maps Charts
Multimdia
Face alignment/recognition
Lighting by guide ambient rendering
Lighting by guide model of the scene
Lighting by guide visual guide by artists
Lighting by guide lighting by lighting artists
Lighting by guide
Lighting by guide
Lighting by guide
Lighting by guide result guide