Created by [ Anita Hu], last modified on May 09, 2020
What happened during W20?
TLDR, much things are done but all need to be integrated and further tested with real data
\
Blockers for S20
\
+—————–+—————–+—————–+—————–+ | Tasks for S20 | Importance | Rough time | Details on task | | | | estimates | | +=================+=================+=================+=================+ | Camera | Critical | 1 week (this | - Found some | | coordinate → | | shouldn't take | code we can | | world | | too long) | use <https: | | coordinate | | | //github.com/sh | | conversion | | | aoxuan92/OpenCV | | | | | -2Dto3D> | | | | | - Verify that | | | | | the code is | | | | | good | | | | | - Get the | | | | | code ready | | | | | for | | | | | integration | | | | | (AKA put it | | | | | into a | | | | | function) | +—————–+—————–+—————–+—————–+ | Integrate all | Critical | 3-4 weeks or | - Test | | components of | | more? | individual | | road line | | | components | | detection into | | | - Write a ROS | | a ROS node | | | node and | | | | | integrate | | | | | functions | | | | | one by one | | | | | - Make sure | | | | | things | | | | | still work | | | | | after it is | | | | | integrated | +—————–+—————–+—————–+—————–+ | Writing the CDR | Critical (CDR | ?? | - Writing the | | (road line | submission due | | report | | detection | July 24) | | (need 4 | | innovative | | | pages) | | concept thing) | | | - Details | | | | | | | | | | <https://do | | | | | cs.google.com/d | | | | | ocument/d/1LnFS | | | | | mRtX-KlztNmIis_ | | | | | rd9zF0mUEbPlGMN | | | | | zx6UO-Fpg/edit? | | | | | folder=0ADRSHK4 | | | | | SUeNNUk9PVA> | +—————–+—————–+—————–+—————–+ | Frustums need | Important but | 2 weeks? | - Changes to | | to support | not urgent | | ROS node on | | detections from | | | multicam | | multiple | | | branch | | cameras | | | - Use | | | | | multipl | | | | | e | | | | | camera | | | | | project | | | | | ion | | | | | matrice | | | | | s | | | | | - Need to | | | | | get rid | | | | | of | | | | | duplica | | | | | te | | | | | detecti | | | | | ons | | | | | because | | | | | cameras | | | | | have | | | | | overlap | | | | | ping | | | | | FOV | +—————–+—————–+—————–+—————–+ | Improvements to | Important but | 2 weeks max | - Better | | traffic light | not urgent | | traffic | | detection | | | light | | | | | bounding | | | | | box | | | | | detection | | | | | by cropping | | | | | the input | | | | | image (ROI | | | | | on right | | | | | camera | | | | | frames) | | | | | - Update code | | | | | for traffic | | | | | light arrow | | | | | signs | +—————–+—————–+—————–+—————–+ | Improvements to | Somewhat | 3-4 weeks or | - We get a | | frustums (AKA | | more? | frustum | | camera and | | | containing | | lidar fusion | | | lidar | | code) | | | points for | | | | | each 2D | | | | | object we | | | | | detect | | | | | - Need to | | | | | find a | | | | | better way | | | | | to get the | | | | | correct 3D | | | | | bounding | | | | | box for | | | | | each object | | | | | inside the | | | | | frustum | | | | | (currently | | | | | we do | | | | | clustering | | | | | inside the | | | | | frustum and | | | | | select the | | | | | 3D bounding | | | | | box based | | | | | on IOU with | | | | | 2D bounding | | | | | box when | | | | | projected | | | | | onto the | | | | | image) | | | | | - Perhaps | | | | | some | | | | | literature | | | | | review | | | | | first | | | | | - Implement | | | | | and test | | | | | with ROS | | | | | bags to | | | | | make sure | | | | | it's | | | | | actually | | | | | better | +—————–+—————–+—————–+—————–+ | Improvements to | Somewhat | 3-4 weeks or | - Some new | | 2D object | | more? | papers are | | detection | | | out, let's | | | | | try some of | | | | | them out | | | | | - YOLOv4 | | | | | - Efficie | | | | | ntDet | | | | | - Models | | | | | trained | | | | | with COCO | | | | | dataset are | | | | | good for | | | | | our use. If | | | | | there are | | | | | pre-trained | | | | | models, | | | | | that would | | | | | be great. | | | | | - Questions | | | | | to answer | | | | | - Accurac | | | | | y | | | | | compare | | | | | d | | | | | to the | | | | | model | | | | | we | | | | | current | | | | | ly | | | | | use | | | | | → YOLOv | | | | | 3 | | | | | 416 | | | | | (weight | | | | | s | | | | | from | | | | | this | | | | | website | | | | | <https://pjred | | | | | die.com/darknet | | | | | /yolo/>) | | | | | - What is | | | | | the FPS | | | | | of | | | | | detecti | | | | | on | | | | | on the | | | | | rugged | | | | | (create | | | | | user on | | | | | rugged | | | | | and ssh | | | | | to | | | | | test) | | | | | - Does it | | | | | work | | | | | with | | | | | OpenVIN | | | | | O? | | | | | (try | | | | | convert | | | | | ing | | | | | to the | | | | | OpenVIN | | | | | O | | | | | format, | | | | | just | | | | | need a | | | | | compute | | | | | r | | | | | with | | | | | OpenVIN | | | | | O | | | | | install | | | | | ed, | | | | | wato | | | | | servers | | | | | would | | | | | work)\ | | | | | - If | | | | | not | | | | | wor | | | | | k, | | | | | if | | | | | the | | | | | FPS | | | | | is | | | | | fas | | | | | t | | | | | whe | | | | | n | | | | | run | | | | | ning | | | | | it | | | | | wit | | | | | h | | | | | pyt | | | | | hon | | | | | on | | | | | the | | | | | rug | | | | | ged | | | | | and | | | | | the | | | | | acc | | | | | uracy | | | | | is | | | | | rea | | | | | lly | | | | | goo | | | | | d, | | | | | the | | | | | n | | | | | we | | | | | can | | | | | con | | | | | sider | | | | | mak | | | | | ing | | | | | a | | | | | pyt | | | | | hon | | | | | ROS | | | | | nod | | | | | e | | | | | for | | | | | thi | | | | | s | +—————–+—————–+—————–+—————–+
Document generated by Confluence on Dec 10, 2021 04:02