Training models for road scene understanding with ... · Automated ground truth / Cross-sensor...
Transcript of Training models for road scene understanding with ... · Automated ground truth / Cross-sensor...
![Page 1: Training models for road scene understanding with ... · Automated ground truth / Cross-sensor learning 1. Solve an easier problem - Run time - Completeness 2. Promise - Scalability](https://reader035.fdocuments.net/reader035/viewer/2022070815/5f0f4be37e708231d443755c/html5/thumbnails/1.jpg)
Training models for road scene understanding with automated ground truth
Dan Levi
With: Noa Garnett, Ethan Fetaya, Shai Silberstein, Rafi Cohen, Shaul Oron, Uri Verner, Ariel Ayash, Kobi Horn, Vlad Golder, Tomer Peer
![Page 2: Training models for road scene understanding with ... · Automated ground truth / Cross-sensor learning 1. Solve an easier problem - Run time - Completeness 2. Promise - Scalability](https://reader035.fdocuments.net/reader035/viewer/2022070815/5f0f4be37e708231d443755c/html5/thumbnails/2.jpg)
GM Advanced Technical Center in Israel (ATCI)
![Page 3: Training models for road scene understanding with ... · Automated ground truth / Cross-sensor learning 1. Solve an easier problem - Run time - Completeness 2. Promise - Scalability](https://reader035.fdocuments.net/reader035/viewer/2022070815/5f0f4be37e708231d443755c/html5/thumbnails/3.jpg)
Agenda
• Road scene understanding
• Acquiring training data with automated ground truth (AGT)
• Test cases:• General obstacle detection• Road segmentation• General obstacle classification• Curb detection• Freespace
• Challenges and limitations
• Summary and future work
![Page 4: Training models for road scene understanding with ... · Automated ground truth / Cross-sensor learning 1. Solve an easier problem - Run time - Completeness 2. Promise - Scalability](https://reader035.fdocuments.net/reader035/viewer/2022070815/5f0f4be37e708231d443755c/html5/thumbnails/4.jpg)
On-board road scene understanding
Static:
• Road edge
• Road markings, complex lane understanding
• Signs
• Obstacles: clutter, construction zone cones
Dynamic:
• Classified objects (cars, pedestrians, bicycles, animals …)
• General obstacles: animals, carts
![Page 5: Training models for road scene understanding with ... · Automated ground truth / Cross-sensor learning 1. Solve an easier problem - Run time - Completeness 2. Promise - Scalability](https://reader035.fdocuments.net/reader035/viewer/2022070815/5f0f4be37e708231d443755c/html5/thumbnails/5.jpg)
Obstacle detection: general and category based
![Page 6: Training models for road scene understanding with ... · Automated ground truth / Cross-sensor learning 1. Solve an easier problem - Run time - Completeness 2. Promise - Scalability](https://reader035.fdocuments.net/reader035/viewer/2022070815/5f0f4be37e708231d443755c/html5/thumbnails/6.jpg)
General obstacles, freespace, road segmentation
- Road segmentation (vision)Mono-camera using semantic segmentation
- General obstacle detection - Freespace (all non-flat road delimiters)3D sensors (Stereo, Lidar)
![Page 7: Training models for road scene understanding with ... · Automated ground truth / Cross-sensor learning 1. Solve an easier problem - Run time - Completeness 2. Promise - Scalability](https://reader035.fdocuments.net/reader035/viewer/2022070815/5f0f4be37e708231d443755c/html5/thumbnails/7.jpg)
Training Data
Revisiting Unreasonable Effectiveness of Data in Deep Learning Era [Sun et al. 2017]
![Page 8: Training models for road scene understanding with ... · Automated ground truth / Cross-sensor learning 1. Solve an easier problem - Run time - Completeness 2. Promise - Scalability](https://reader035.fdocuments.net/reader035/viewer/2022070815/5f0f4be37e708231d443755c/html5/thumbnails/8.jpg)
Manual Annotation
• Time: ~60 min per image
• ~1000 annotators
The Cityscapes Dataset for Semantic Urban Scene Understanding [Cordts et al. 2016]
![Page 9: Training models for road scene understanding with ... · Automated ground truth / Cross-sensor learning 1. Solve an easier problem - Run time - Completeness 2. Promise - Scalability](https://reader035.fdocuments.net/reader035/viewer/2022070815/5f0f4be37e708231d443755c/html5/thumbnails/9.jpg)
Computer graphics simulated data
• Photo-realism
• Scenario generation
![Page 10: Training models for road scene understanding with ... · Automated ground truth / Cross-sensor learning 1. Solve an easier problem - Run time - Completeness 2. Promise - Scalability](https://reader035.fdocuments.net/reader035/viewer/2022070815/5f0f4be37e708231d443755c/html5/thumbnails/10.jpg)
Simulated data for optical flow
FlowNet: Learning Optical Flow with Convolutional Networks [Dosovitskiy et. al. 2015]
![Page 11: Training models for road scene understanding with ... · Automated ground truth / Cross-sensor learning 1. Solve an easier problem - Run time - Completeness 2. Promise - Scalability](https://reader035.fdocuments.net/reader035/viewer/2022070815/5f0f4be37e708231d443755c/html5/thumbnails/11.jpg)
Automated ground truth(AGT) / Cross-sensor learning
Velodyne LIDAR
![Page 12: Training models for road scene understanding with ... · Automated ground truth / Cross-sensor learning 1. Solve an easier problem - Run time - Completeness 2. Promise - Scalability](https://reader035.fdocuments.net/reader035/viewer/2022070815/5f0f4be37e708231d443755c/html5/thumbnails/12.jpg)
Depth from a single image
Semi-Supervised Deep Learning for Monocular Depth Map Prediction [Kuznietsov et al. 2017]
![Page 13: Training models for road scene understanding with ... · Automated ground truth / Cross-sensor learning 1. Solve an easier problem - Run time - Completeness 2. Promise - Scalability](https://reader035.fdocuments.net/reader035/viewer/2022070815/5f0f4be37e708231d443755c/html5/thumbnails/13.jpg)
Map supervised road detection
Map-supervised road detection [Laddha et al. 2016]
![Page 14: Training models for road scene understanding with ... · Automated ground truth / Cross-sensor learning 1. Solve an easier problem - Run time - Completeness 2. Promise - Scalability](https://reader035.fdocuments.net/reader035/viewer/2022070815/5f0f4be37e708231d443755c/html5/thumbnails/14.jpg)
AGT for road scene understanding – general setup
“Target” sensors
Perquisite: Full alignment and synchronization between sensors
“Supervising” sensors
![Page 15: Training models for road scene understanding with ... · Automated ground truth / Cross-sensor learning 1. Solve an easier problem - Run time - Completeness 2. Promise - Scalability](https://reader035.fdocuments.net/reader035/viewer/2022070815/5f0f4be37e708231d443755c/html5/thumbnails/15.jpg)
AGT for road scene understanding: scheme
“Supervising” sensors:
“Target” sensors:
Task: object detection
Data
AGT:1. Compute Task on Supervising sensors:- Offline- Temporal
Ground truth
AGT:2. Project output to target sensor domain
![Page 16: Training models for road scene understanding with ... · Automated ground truth / Cross-sensor learning 1. Solve an easier problem - Run time - Completeness 2. Promise - Scalability](https://reader035.fdocuments.net/reader035/viewer/2022070815/5f0f4be37e708231d443755c/html5/thumbnails/16.jpg)
Automated ground truth / Cross-sensor learning
1. Solve an “easier” problem- Run time- Completeness
2. Promise- Scalability- Continuous (un-bounded)
improvement
1. Challenging setup2. Annotation quality / accuracy3. Inherent limitations of “supervisor”:- Learning beyond supervisor
capabilities- Learning from the same sensor
(bootstrapping)
![Page 17: Training models for road scene understanding with ... · Automated ground truth / Cross-sensor learning 1. Solve an easier problem - Run time - Completeness 2. Promise - Scalability](https://reader035.fdocuments.net/reader035/viewer/2022070815/5f0f4be37e708231d443755c/html5/thumbnails/17.jpg)
AGT for General obstacle detection
“Supervising” sensors:
“Target” sensors:
Task: General obs. Det.
Data
Ground truthAGT
Velodyne HDL64
Front camera
![Page 18: Training models for road scene understanding with ... · Automated ground truth / Cross-sensor learning 1. Solve an easier problem - Run time - Completeness 2. Promise - Scalability](https://reader035.fdocuments.net/reader035/viewer/2022070815/5f0f4be37e708231d443755c/html5/thumbnails/18.jpg)
StixelNet: Monocular obstacle detection
Levi, Dan, Noa Garnett, Ethan Fetaya. StixelNet : A Deep Convolutional Network for Obstacle Detection and Road Segmentation. In BMVC 2015.
![Page 19: Training models for road scene understanding with ... · Automated ground truth / Cross-sensor learning 1. Solve an easier problem - Run time - Completeness 2. Promise - Scalability](https://reader035.fdocuments.net/reader035/viewer/2022070815/5f0f4be37e708231d443755c/html5/thumbnails/19.jpg)
StixelNet column based approach
INPUT
OUTPUT
![Page 20: Training models for road scene understanding with ... · Automated ground truth / Cross-sensor learning 1. Solve an easier problem - Run time - Completeness 2. Promise - Scalability](https://reader035.fdocuments.net/reader035/viewer/2022070815/5f0f4be37e708231d443755c/html5/thumbnails/20.jpg)
AGT for obstacle detection – version I
KITTI Dataset [Geiger et al. 2013]
![Page 21: Training models for road scene understanding with ... · Automated ground truth / Cross-sensor learning 1. Solve an easier problem - Run time - Completeness 2. Promise - Scalability](https://reader035.fdocuments.net/reader035/viewer/2022070815/5f0f4be37e708231d443755c/html5/thumbnails/21.jpg)
Velodyne LIDAR
AGT for obstacle detection – version I
• Raw images: 56 sequences (50 Train, 6 Test). • 6,000 train images (every 5th frame) and 800 test.• Ground truth result:• After GT: 331K training columns and 57K testing
![Page 22: Training models for road scene understanding with ... · Automated ground truth / Cross-sensor learning 1. Solve an easier problem - Run time - Completeness 2. Promise - Scalability](https://reader035.fdocuments.net/reader035/viewer/2022070815/5f0f4be37e708231d443755c/html5/thumbnails/22.jpg)
AGT for general obstacles in image plane
Figure taken from [Fernandes et al. 2015]
1. Project Lidar points to image plane 2. Interpolate depth to all pixels
3. Find columns with depth profile typical to transition: road roughly vertical obstacle
![Page 23: Training models for road scene understanding with ... · Automated ground truth / Cross-sensor learning 1. Solve an easier problem - Run time - Completeness 2. Promise - Scalability](https://reader035.fdocuments.net/reader035/viewer/2022070815/5f0f4be37e708231d443755c/html5/thumbnails/23.jpg)
General obstacles AGT examples
Limitations:- Cannot handle: close obstacles, ‘’clear’’ columns- Low coverage (~30%)
![Page 24: Training models for road scene understanding with ... · Automated ground truth / Cross-sensor learning 1. Solve an easier problem - Run time - Completeness 2. Promise - Scalability](https://reader035.fdocuments.net/reader035/viewer/2022070815/5f0f4be37e708231d443755c/html5/thumbnails/24.jpg)
INPUT OUTPUT
64200
5
3
Max pooling (8X4)
45
5
11
1
1024
Dense Dense Dense
Layer 1:convolution
al
Layer 2:convolution
al
Layer 3: fully connected
Layer 4:fully
connected
Layer 5:fully
connected
11
5
24
370
2048
50
3
Max pooling (4X3)
StixelNet (v1) 5 Layer CNN
y
![Page 25: Training models for road scene understanding with ... · Automated ground truth / Cross-sensor learning 1. Solve an easier problem - Run time - Completeness 2. Promise - Scalability](https://reader035.fdocuments.net/reader035/viewer/2022070815/5f0f4be37e708231d443755c/html5/thumbnails/25.jpg)
![Page 26: Training models for road scene understanding with ... · Automated ground truth / Cross-sensor learning 1. Solve an easier problem - Run time - Completeness 2. Promise - Scalability](https://reader035.fdocuments.net/reader035/viewer/2022070815/5f0f4be37e708231d443755c/html5/thumbnails/26.jpg)
Experimental Results (Max Probability)
![Page 27: Training models for road scene understanding with ... · Automated ground truth / Cross-sensor learning 1. Solve an easier problem - Run time - Completeness 2. Promise - Scalability](https://reader035.fdocuments.net/reader035/viewer/2022070815/5f0f4be37e708231d443755c/html5/thumbnails/27.jpg)
Stereo [Badino et al. 2009]“StixelNet”
Comparison with stereo
![Page 28: Training models for road scene understanding with ... · Automated ground truth / Cross-sensor learning 1. Solve an easier problem - Run time - Completeness 2. Promise - Scalability](https://reader035.fdocuments.net/reader035/viewer/2022070815/5f0f4be37e708231d443755c/html5/thumbnails/28.jpg)
AGT for Obstacle classification
“Supervising” sensors:
“Target” sensors:
Task: Obstacle classification
Data
Ground truthAGT
Front camera
![Page 29: Training models for road scene understanding with ... · Automated ground truth / Cross-sensor learning 1. Solve an easier problem - Run time - Completeness 2. Promise - Scalability](https://reader035.fdocuments.net/reader035/viewer/2022070815/5f0f4be37e708231d443755c/html5/thumbnails/29.jpg)
AGT for obstacle classification
Image based detection
Source: http://self-driving-future.com/the-eyes/velodyne/
Lidar based verification
![Page 30: Training models for road scene understanding with ... · Automated ground truth / Cross-sensor learning 1. Solve an easier problem - Run time - Completeness 2. Promise - Scalability](https://reader035.fdocuments.net/reader035/viewer/2022070815/5f0f4be37e708231d443755c/html5/thumbnails/30.jpg)
Obstacle classification trained net result: pedestrians
![Page 31: Training models for road scene understanding with ... · Automated ground truth / Cross-sensor learning 1. Solve an easier problem - Run time - Completeness 2. Promise - Scalability](https://reader035.fdocuments.net/reader035/viewer/2022070815/5f0f4be37e708231d443755c/html5/thumbnails/31.jpg)
AGT for General obstacle detection (ver 2)
“Supervising” sensors:
“Target” sensors:
Task: General obs. Det.
Data
Ground truthAGT
Velodyne HDL64
Front camera
![Page 32: Training models for road scene understanding with ... · Automated ground truth / Cross-sensor learning 1. Solve an easier problem - Run time - Completeness 2. Promise - Scalability](https://reader035.fdocuments.net/reader035/viewer/2022070815/5f0f4be37e708231d443755c/html5/thumbnails/32.jpg)
Unified network: StixelNet + Object detection + Object pose estimation
Noa Garnett, Shai Silberstein, Shaul Oron, Ethan Fetaya, Uri Verner, Ariel Ayash, Vlad Goldner, Rafi Cohen, Kobi Horn, Dan Levi. Real-time category-based and general obstacle detection for autonomous driving. CVRSUAD Workshop, ICCV2017.
![Page 33: Training models for road scene understanding with ... · Automated ground truth / Cross-sensor learning 1. Solve an easier problem - Run time - Completeness 2. Promise - Scalability](https://reader035.fdocuments.net/reader035/viewer/2022070815/5f0f4be37e708231d443755c/html5/thumbnails/33.jpg)
Object-centric obstacle detection AGTEstimate and subtract road
plane
3D Clustering (objects above
20cm)
Project to image, smooth
Bottom contour via dynamic
programming
Detect clear columns:No object above 5cm + far enough returns
‘’near’’ obstacles:1. Below lidar
coverage2. During training
![Page 34: Training models for road scene understanding with ... · Automated ground truth / Cross-sensor learning 1. Solve an easier problem - Run time - Completeness 2. Promise - Scalability](https://reader035.fdocuments.net/reader035/viewer/2022070815/5f0f4be37e708231d443755c/html5/thumbnails/34.jpg)
General obstacles: old vs. new AGT
![Page 35: Training models for road scene understanding with ... · Automated ground truth / Cross-sensor learning 1. Solve an easier problem - Run time - Completeness 2. Promise - Scalability](https://reader035.fdocuments.net/reader035/viewer/2022070815/5f0f4be37e708231d443755c/html5/thumbnails/35.jpg)
New general obstacle dataset with fisheye lens camera
#images #instances (columns)
Kitti--train 6K 5M
Internal-train
16K 20M
Kitti-test 760 11K
Internal-test 910 19K
![Page 36: Training models for road scene understanding with ... · Automated ground truth / Cross-sensor learning 1. Solve an easier problem - Run time - Completeness 2. Promise - Scalability](https://reader035.fdocuments.net/reader035/viewer/2022070815/5f0f4be37e708231d443755c/html5/thumbnails/36.jpg)
StixelNet2: New network architecture
![Page 37: Training models for road scene understanding with ... · Automated ground truth / Cross-sensor learning 1. Solve an easier problem - Run time - Completeness 2. Promise - Scalability](https://reader035.fdocuments.net/reader035/viewer/2022070815/5f0f4be37e708231d443755c/html5/thumbnails/37.jpg)
Improved results on KITTI
![Page 38: Training models for road scene understanding with ... · Automated ground truth / Cross-sensor learning 1. Solve an easier problem - Run time - Completeness 2. Promise - Scalability](https://reader035.fdocuments.net/reader035/viewer/2022070815/5f0f4be37e708231d443755c/html5/thumbnails/38.jpg)
0
0.2
0.4
0.6
0.8
1
Kitti - max Pr. Internal - max Pr. kitti - avg. Pr Internal - avg. Pr
Old New
Experimental results with new AGT
![Page 39: Training models for road scene understanding with ... · Automated ground truth / Cross-sensor learning 1. Solve an easier problem - Run time - Completeness 2. Promise - Scalability](https://reader035.fdocuments.net/reader035/viewer/2022070815/5f0f4be37e708231d443755c/html5/thumbnails/39.jpg)
Experimental results with new AGT
0
0.2
0.4
0.6
0.8
1
Kitti - max Pr. Internal - max Pr. kitti - avg. Pr Internal - avg. Pr
Chart TitleOld New
Edge cases excluded (“near”, “clear”)
![Page 40: Training models for road scene understanding with ... · Automated ground truth / Cross-sensor learning 1. Solve an easier problem - Run time - Completeness 2. Promise - Scalability](https://reader035.fdocuments.net/reader035/viewer/2022070815/5f0f4be37e708231d443755c/html5/thumbnails/40.jpg)
Cross dataset generalization
0
0.2
0.4
0.6
0.8
1
kitti-test max kitti-test avg internal-testmax
internal-testavg
Train on KITTI Train on internal Train on both
![Page 41: Training models for road scene understanding with ... · Automated ground truth / Cross-sensor learning 1. Solve an easier problem - Run time - Completeness 2. Promise - Scalability](https://reader035.fdocuments.net/reader035/viewer/2022070815/5f0f4be37e708231d443755c/html5/thumbnails/41.jpg)
AGT for car pose estimation
“Supervising” sensors:
“Target” sensors:
Task: pose estimation
Data
Ground truthAGT
IMU
![Page 42: Training models for road scene understanding with ... · Automated ground truth / Cross-sensor learning 1. Solve an easier problem - Run time - Completeness 2. Promise - Scalability](https://reader035.fdocuments.net/reader035/viewer/2022070815/5f0f4be37e708231d443755c/html5/thumbnails/42.jpg)
AGT for pose estimation
Source: http://self-driving-future.com/the-eyes/velodyne/
Multi sensor, temporal object detection
8 orientation bins pose representation
Dynamic Static
![Page 43: Training models for road scene understanding with ... · Automated ground truth / Cross-sensor learning 1. Solve an easier problem - Run time - Completeness 2. Promise - Scalability](https://reader035.fdocuments.net/reader035/viewer/2022070815/5f0f4be37e708231d443755c/html5/thumbnails/43.jpg)
Pose estimation
trained with mixed AGT and Manual
![Page 44: Training models for road scene understanding with ... · Automated ground truth / Cross-sensor learning 1. Solve an easier problem - Run time - Completeness 2. Promise - Scalability](https://reader035.fdocuments.net/reader035/viewer/2022070815/5f0f4be37e708231d443755c/html5/thumbnails/44.jpg)
AGT for Curb detection
![Page 45: Training models for road scene understanding with ... · Automated ground truth / Cross-sensor learning 1. Solve an easier problem - Run time - Completeness 2. Promise - Scalability](https://reader035.fdocuments.net/reader035/viewer/2022070815/5f0f4be37e708231d443755c/html5/thumbnails/45.jpg)
Curb detection trained net result examples
![Page 46: Training models for road scene understanding with ... · Automated ground truth / Cross-sensor learning 1. Solve an easier problem - Run time - Completeness 2. Promise - Scalability](https://reader035.fdocuments.net/reader035/viewer/2022070815/5f0f4be37e708231d443755c/html5/thumbnails/46.jpg)
AGT for freespace
“Supervising” sensors:
“Target” sensors:
Task: freespace
Data
Ground truthAGT
![Page 47: Training models for road scene understanding with ... · Automated ground truth / Cross-sensor learning 1. Solve an easier problem - Run time - Completeness 2. Promise - Scalability](https://reader035.fdocuments.net/reader035/viewer/2022070815/5f0f4be37e708231d443755c/html5/thumbnails/47.jpg)
AGT for freespace with 3D beams
Analyze single Lidar “Beam”
Estimate and subtract road plane
Project freespace limit to image plane, find ‘’near’’ and ‘’clear’’
Project limit to ground plane
Velodyne
Velodyne scan direction
![Page 48: Training models for road scene understanding with ... · Automated ground truth / Cross-sensor learning 1. Solve an easier problem - Run time - Completeness 2. Promise - Scalability](https://reader035.fdocuments.net/reader035/viewer/2022070815/5f0f4be37e708231d443755c/html5/thumbnails/48.jpg)
Obstacles vs. Freespace AGT
![Page 49: Training models for road scene understanding with ... · Automated ground truth / Cross-sensor learning 1. Solve an easier problem - Run time - Completeness 2. Promise - Scalability](https://reader035.fdocuments.net/reader035/viewer/2022070815/5f0f4be37e708231d443755c/html5/thumbnails/49.jpg)
Freespace + object detection + car 3D pose
![Page 50: Training models for road scene understanding with ... · Automated ground truth / Cross-sensor learning 1. Solve an easier problem - Run time - Completeness 2. Promise - Scalability](https://reader035.fdocuments.net/reader035/viewer/2022070815/5f0f4be37e708231d443755c/html5/thumbnails/50.jpg)
Freespace + object detection + car 3D pose
![Page 51: Training models for road scene understanding with ... · Automated ground truth / Cross-sensor learning 1. Solve an easier problem - Run time - Completeness 2. Promise - Scalability](https://reader035.fdocuments.net/reader035/viewer/2022070815/5f0f4be37e708231d443755c/html5/thumbnails/51.jpg)
Freespace + object detection + car 3D pose
![Page 52: Training models for road scene understanding with ... · Automated ground truth / Cross-sensor learning 1. Solve an easier problem - Run time - Completeness 2. Promise - Scalability](https://reader035.fdocuments.net/reader035/viewer/2022070815/5f0f4be37e708231d443755c/html5/thumbnails/52.jpg)
Freespace + object detection + car 3D pose
![Page 53: Training models for road scene understanding with ... · Automated ground truth / Cross-sensor learning 1. Solve an easier problem - Run time - Completeness 2. Promise - Scalability](https://reader035.fdocuments.net/reader035/viewer/2022070815/5f0f4be37e708231d443755c/html5/thumbnails/53.jpg)
Freespace + object detection + car 3D pose
![Page 54: Training models for road scene understanding with ... · Automated ground truth / Cross-sensor learning 1. Solve an easier problem - Run time - Completeness 2. Promise - Scalability](https://reader035.fdocuments.net/reader035/viewer/2022070815/5f0f4be37e708231d443755c/html5/thumbnails/54.jpg)
Finetuning from AGT: road segmentation
1. Fine-tune on KITTI Road segmentation (manually labelled)
2. Graph-cut segmentation
3. State-of-the-art accuracy among non-anonymous (94.88% MaxF)
![Page 55: Training models for road scene understanding with ... · Automated ground truth / Cross-sensor learning 1. Solve an easier problem - Run time - Completeness 2. Promise - Scalability](https://reader035.fdocuments.net/reader035/viewer/2022070815/5f0f4be37e708231d443755c/html5/thumbnails/55.jpg)
AGT challenges: How accurate is the AGT?
![Page 56: Training models for road scene understanding with ... · Automated ground truth / Cross-sensor learning 1. Solve an easier problem - Run time - Completeness 2. Promise - Scalability](https://reader035.fdocuments.net/reader035/viewer/2022070815/5f0f4be37e708231d443755c/html5/thumbnails/56.jpg)
AGT challenges: calibration, synchronization
![Page 57: Training models for road scene understanding with ... · Automated ground truth / Cross-sensor learning 1. Solve an easier problem - Run time - Completeness 2. Promise - Scalability](https://reader035.fdocuments.net/reader035/viewer/2022070815/5f0f4be37e708231d443755c/html5/thumbnails/57.jpg)
AGT Perception mistakes
Non-flat road
Assumptions / coverage
![Page 58: Training models for road scene understanding with ... · Automated ground truth / Cross-sensor learning 1. Solve an easier problem - Run time - Completeness 2. Promise - Scalability](https://reader035.fdocuments.net/reader035/viewer/2022070815/5f0f4be37e708231d443755c/html5/thumbnails/58.jpg)
Thank you!