Accelerate your neural networks today!
ENOT.ai - the only AI optimization tool
for automotive industry
for automotive industry
- up to 8 times
neural network
acceleration - up to 10 times
neural network
compression - up to 70%
hardware cost
reduction - ~0.00%
accuracy
trade-off
Value Proposition
- 1Seamless On-Device Integration of Language Models into Chips or Vehicles, Eliminating Cloud Dependency
- 2Accelerated Time-to-Market with New Feature Introductions
- 3New Revenue Streams Unlocked Through Personalization
- 4Enhanced Interaction with Language Models for a Better User Experience
- 5Safer, Faster, and More Stable Decision-Making
- 6Enabling New Features, Such as Driver Coaching
- 7Autonomous driving and ADAS Features Enablement
ENOT.ai – tool for Automotive AI developers
- Timing
- Budget
- Success
Baseline
Baseline
Data collection
Porting
80% time spent
Baseline
Data collection
Baseline
Data collection
Test
NN train
Hypothesis
Timing ⠀⠀ Budget⠀ ⠀ Success
Technology
ENOT.ai works within customers infrastructure, so the companies do not need to share their super confidential data with 3rd parties or send it to cloud infrastructure if they want to keep it within the organisation
Step 1
Step 2
- Upload Neural Network
- Upload Dataset
Result
- ENOT accelerate / compress NN
- ENOT fine-tune NN on your data
- 3x+ higher efficiency
- 3x+ lower latency of AI
- ~0% accuracy loss
Step 1
Step 2
Step 3
ENOT.ai - tool for Automotive AI developers
Curiosity about life in all its aspects, I think, is still the secret of great creative people.
L = L
+
task
latency
a*L
Core applications
ENOT.ai software can accelerate neural networks by 3x-8x without loss of accuracy for any AI chip
- ADAS Autonomous vehicles
- Secure Car Access
- Vehicle AI-assistant
- Driver safety monitoring
⠀
⠀
⠀⠀
Supported platforms
NN model format
NN types
Pb, tf-lite, ONNX
LSTM CNN
DNN RNN
DNN RNN
Deployment
Runtime
On premises
On the cloud
On the cloud
Hardware libraries
GPU CPU
NPU FPGA
NPU FPGA
+ many more
Machine learning
frameworks
frameworks
Case Studies: Optimizing Autopilot Performance in a Leading Open Source Project
Making plane prediction perform 4.3x times faster:
Before:
After:
MACs: 472774912
Acceleration: 1.0x
MACs: 109917744
Acceleration: 4.3x
Case Studies: Enhancing Advanced Driver Assistance Systems for a Leading Global Automotive Technology Provider
Task: Improve lane detection speed on target hardware platform for Advanced Driver Assistance System (ADAS) used in company's autonomous driving solutions.
Problem: real-time processing with certain level of accuracy is critical for ADAS.
Solution: ENOT reduced neural network inference latency by 3x times for target hardware platform.
Results: 3x faster reaction in case of driving out of the lane.
Problem: real-time processing with certain level of accuracy is critical for ADAS.
Solution: ENOT reduced neural network inference latency by 3x times for target hardware platform.
Results: 3x faster reaction in case of driving out of the lane.
- Client's expectation: 2x acceleration
- ENOT.ai achievement: 3x acceleration
Results:
Case Studies: Enhancing Performance for a Global Consumer Electronics Leader
ENOT.ai has improved the speed of clients's ADAS (Advanced Driver Assistance System).
Task: In this case, neural networks were being used to segment the input images and recognize objects in the video, and the requirement was to speed up video processing without loss of accuracy while keeping the same neural network architecture (Unet) and hardware (Nvidia Jetson TX2).
Task: In this case, neural networks were being used to segment the input images and recognize objects in the video, and the requirement was to speed up video processing without loss of accuracy while keeping the same neural network architecture (Unet) and hardware (Nvidia Jetson TX2).
- acceleration 1.97x
- 6.40x without limitation
Results:
Model
IoU
Latency
Acceliration
Baseline
enot.ai
enot.ai w/o limitations
79.98
79.15
~79.15
33.1 ms/image
16.79 ms/image
~5.17 ms/image
1.00
1.79
~6.40
Further improvements (limitations that could be removed):
- By choosing different hardware, additional 1.3x acceleration could have been achieved;
- By changing input resolution, another 1.25x acceleration could have been achieved;
- Our software’s recent update has addressed skip-connection related to Unet, thus granting another 2.00x acceleration.
Case Studies: Advancing Agricultural Efficiency with Laser Weeding Technology
The client is a company that develops laser weeding machinery. This machinery applies laser beams to weeds, thus boiling the water in the leaves. This approach is more productive than manual weeding and more environmentally friendly than herbicides.
Task: The baseline segmentation model used high resolution RGB images to be able to detect the smallest parts of the plants. Client set two objectives:
Task: The baseline segmentation model used high resolution RGB images to be able to detect the smallest parts of the plants. Client set two objectives:
- Accelerate the model speed by at least 2x with an accuracy loss of no more than 0.02 (quality metric - eIoU mAP@[.5:.95]).
- enot.ai should suggest model parameters to achieve a higher recognition accuracy.
- acceleration 2.72x
- accuracy increase 25%
Results:
Model
Accuracy (eloU mAP @[.5:.95])
Latency
Acceliration
Client's Baseline
enot.ai v1
0.5321
0.5521
0.6628
22.0±0.57
8.08±0.5
14.44±0.2
2.72x
1.52x
Accuracy Increase
enot.ai v2
Baseline
Baseline
1.04x
1.25x
Case Studies: Enhancing Privacy in Video Technology for a Premier Electronics Manufacturer
Task: Face bluring for GDPR compliant camera
Problem: Chipset has only 8 Mb RAM. In Order to run “standard” neural network it’s required at least 34 Mb RAM
Solution: ENOT team with help of ENOT’s neural network technology developed a highly accurate solution for 8Mb chipset to blur faces on full HD video in real-time
Problem: Chipset has only 8 Mb RAM. In Order to run “standard” neural network it’s required at least 34 Mb RAM
Solution: ENOT team with help of ENOT’s neural network technology developed a highly accurate solution for 8Mb chipset to blur faces on full HD video in real-time
