An Open Source Approach to Image/Video Machine Learning Sponsored Content by AMD
Image and video machine learning capabilities are highly desirable in a wide variety of application areas including autonomous vehicles, security and surveillance, medical diagnostics, scientific research, and more.
While each application has its own requirements, increasingly, there is growing interest in open source machine learning frameworks including the work being done by the Github community on ROCm, the first open-source HPC/hyperscale-class platform for GPU computing that’s programming-language independent. This work complements other industry open source efforts in deploying research and production quality machine learning applications using the deep learning frameworks PyTorch with Caffe2, TensorFlow, and MIOpen, the open-source deep learning library for AMD GPUs.
The open source nature of these efforts is critical to maintaining the fast-paced developments in the field. To accomplish that, developers of these platforms and frameworks have made them easily accessible. For example, by using APIs and other methods, companies can use ROCm while retaining their investment in existing applications and programming done using CUDA proprietary software.
Technology that powers image and video machine learning
Machine learning applications rely on computer hardware that can support the highest processing capabilities to manage complex data sets from multiple input streams simultaneously.
The fastest way to get to results is to partner with a company that has complete (hardware and software) solutions and real-world expertise in deploying those solutions. This is where AMD can help.
AMD’s machine learning systems deliver high-performance compute (both CPUs and GPUs) with an open ecosystem for software development. AMD’s suite of GPU hardware and open-source software offerings are designed to dramatically increase performance, efficiency, and ease of implementation of machine learning workloads.
On the software front, AMD has been driving the development of ROCm for machine learning. Complementing the software efforts, AMD hardware offers synergistic benefits for machine learning applications. Its Radeon Instinct™ MI Series of accelerators for machine learning is designed to be open from the metal forward. Higher levels of performance and efficiencies are enabled through AMD’s world-class GPU technologies and the Radeon Instinct’s open ecosystem approach via its ROCm software platform and work with open industry standards such as next gen interconnect technologies.
AMD technology is being used in several leading-edge machine learning projects.
One effort is AMD’s work with Highwai simulators, which seek to improve self-driving cars using realistic simulated worlds and environments. The simulators provide the ability to create scenarios too dangerous for real cars and drivers to participate in, such as emergency braking in front of pedestrians, collisions, rollovers, fires, explosions, and accident avoidance. Training a simulator and then using the simulator to make critical decisions requires the use of many GPUs in a balanced system.
Another AMD machine learning effort focused on image segmentation in Advanced Driver Assistance Systems (ADAS). Here, machine learning-based image analysis is conducted on video streams. Again, the work requires many GPUs in a balanced system.
Additional demonstrations of the image and video machine learning analysis work being done using AMD hardware, ROCm, and open source frameworks include:
- Live object detection: https://youtu.be/QGP9aW-Gneo
- Live person detection: https://youtu.be/AE75_PRcpXQ
- Valossa-Genesis video intelligence: https://youtu.be/0l9_VdSEkDw
- Video activity detection: https://youtu.be/VhKrZngAeWA
- Image classification demo: https://youtu.be/CS7yyuQv4jg
The bottom line is that partnering with AMD speeds the time to results in any machine learning project. Using Radeon Instinct™ powered ready-to-deploy solutions for machine learning accelerate project deployments with:
- Easier server deployments
- ROCm Open eCosystem including optimized framework libraries
- Deep learning framework docker containers
- Faster times to application development
For more information about open source machine learning frameworks and GPU accelerated image and video analysis machine learning, visit the following:
ROCm, a New Era in Open GPU Computing
https://rocm.github.io/ or https://rocm-documentation.readthedocs.io/en/latest/index.html
ROCm core technology
https://github.com/RadeonOpenCompute
Radeon Instinct™ products
https://www.amd.com/en/graphics/servers-radeon-instinct-mi
Radeon Instinct deep learning solutions
https://www.amd.com/en/graphics/servers-radeon-instinct-deep-learning
Preparing a machine to run with ROCm and docker
https://github.com/RadeonOpenCompute/ROCm-docker/blob/master/quick-start.md
Preparing a machine to run TensorFlow with docker
https://hub.docker.com/r/rocm/tensorflow/
