fhict-s-koen / litr Goto Github PK

• Mail potential experts
• Write down question
• Meet with expert

Acceptance Criteria

• Example criteria

Test cases

• TC01

Tasks

• Prepare according data

• Add an Agile strategy to the project plan that takes in account the problems of machine learning.
• Add the following labels:
  • user story
  • story points [1,2,3,5,8]
  • spike
  • task
  • sprint [1,2,3,4,5]
• Make a definition of done
• Validate the definition of done
• Define the product goal
• Validate the product goal with the stakeholder
• Define the sprint goal (pretty much define the simplest product is we can deliver also called a MVP)
• Validate the sprint goal with the stakeholder

Tasks

• Answer research question: What datasets regarding litter are already out there?

Learning goal
Is this dataset usable for us and in what way?
Result: We can use this dataset for testing and after labelling, (by using a trained model to label), we can use it for further training (especially for future use of drones).

• Goal: Find an initial algorithm to use for the MVP.

Learning goal
We want to know how to train a model with the YOLO algorithm.

Tasks

• Train model on TACO
• Train model on custom dataset(e.g: TACO and Drinking Waste combined)

Learning goal(s)
Goal: We want to find out how to get the location of the sensor.
Result(s): ...

Task(s)

• Find a GPS-module that has the ability to get the location of the sensor.
• Getting location in code, ready for post.

Learning goal
Is the dataset useable for us and in what way?

Result:
Yes, Drinking waste is split up into 4 categories; aluminium cans(AluCan), plastic bottles (PET), Plastic milk bottles(HDPEM) and Glass bottles(Glass.)
Every image has a .txt file with the corresponding bounding boxes.

Learning goal(s)
Goal: We want to know what the stakeholder wants in the advising document.
Result: ...

Task(s)

• Make a template
• Ask the stakeholder
• Add advice about findings

Think of a way how to acquire the data like:
Drone, Car, Bike Etc.

Learning goal
We want to figure what the NAP framework can be used for.

Tasks

• Look up what NAP is
• Write down possible use cases

Acceptance Criteria

• Must have an autocomplete for municipalities (in the Netherlands)
• Must pan over to the corresponding municipality

Test cases

• TC01 Type in municipality

Tasks

• Add search bar to the geographical-map
• Add code for autocomplete for the municipalities
• Add code to pan over to the corresponding municipality

Acceptance Criteria

• Markers should always be accessible by the user.

Test cases

• TC01 click a cluster of markers
• TC02 click marker
• TC03 click draw bounding box

Tasks

• install leaflet.markercluster
• implement clustering
• fix issues

Acceptance Criteria

• Example criteria

Test cases

• TC01

Tasks

• Prepare according data

Learning goal(s)
Goal: We want to find out if there are heatmap libraries that fit our use case.
Result: ...

Task(s)

• Look up libraries
• Test the best fitting libraries
• Determine what library to use or if we should make something ourselves

We test whether it is possible to run the Yolov5 model on the Jetson Nano.

Acceptance Criteria

• Must test using the GPU (CUDA) of the Jetson.

Test cases

• TC01 Test model Yolov5s
• TC01 Test model Yolov5n

Tasks

• Set up Dockerfile with correct libraries/modules
• Set up real-time object detection script
• Test Yolov5 models on Jetson Nano

• Find an initial object(s) to detect for a prototype.
• Find out what is the most common and harmful litter.

Learning goal:
Have a dataset prepared for our model to use.

Tasks

• Select the datasets to use from:
• TACO - (going to use)
• #23
• #24
• #25 - (going to use)
• Prepare the datasets
• Merge the datasets

Learning goal(s)
Goal: We want to determine what output data is important for us, and how to get it in the correct format.
Result:
Determine what output data is important to us:

• image frames (bitmap array)
• accuracy of model
• type of litter
• time of detection
• location
• number of detections

Task(s)

• Determine what output data is important to us
• Generate all the output data
• Correctly format output data

Acceptance Criteria

• Geographical-map must be able to contain markers
• The user must be able to pan across the map
• The user can zoom in/out the map
• The user can click markers, which contain litter data.

Test case

• Test performance with 1000 markers in the Netherlands
• Test performance with 100 markers per municipality

Tasks

• Add Geographical-map

• Add code to load markers

• Write script to randomly generate 1000 markers in the Netherlands

• Write script to randomly generate 100 markers in each municipality

Goal: Analyze the dataset and provide a visual representation

Results: This dataset can be used to further train and increase the accuracy of an already created machine learning model.

Acceptance Criteria

• Go to current location on button click

Test cases

• TC01 click on current location button

Tasks

• Add code to go back to current based on geolocation or GPS
• Add fallback code if neither work, which sends the user to the latest typed in location

Acceptance Criteria

• Dark mode should reduce strain on eyes.

Test cases

• TC01 Click toggle button for dark/light mode.

Tasks

• Set up code for toggling between dark and light mode.

Learning goals

• We want to know how long it takes to train the model so we can figure out for the future what we're capable of.
• We want to know what hardware is required to run the model.

After training with the different YOLO algorithms:
(1070 ti)
yolov5n: 30s/epoch
yolov5s: 60s/epoch
yolov5m: 90s/epoch
yolov5l: 120s/epoch
yolov5xl: 180s/epoch

Needed hardware:
Jetson with at least 7gb or more ram

Tasks

• determine the estimated training time for our hardware.
• determine what hardware is needed to run the AI at a certain fps.

Scoring based on accuracy & time

Learning goal
Figure out what the best (pre-trained) YOLO algorithms is for us to use.

Tasks

• List of all available YOLO algorithms
• Determine which algorithm to test
• Compare algorithms
• Pick optimal algorithm

Learning goal
We want to know what requirements the heatmap will have.

Tasks

• Write down possible requirements for the heatmap.
• Validate the requirements with the stakeholder.

Acceptance Criteria

• Should show markers and/or clusters on map with corresponding time of the slider.

Test cases

• TC01 Test detections from last 24 hours
• TC02 Test detections from last 3 days
• TC03 Test detections from last 1 week
• TC04 Test detections from all time

Tasks

• Add time slider control
• Add time filter

Acceptance Criteria

• Adding data must require authorization.

Test cases (fix this)

• TC01 Check authorization
• TC02 Check error handling

Tasks

• #51
• #64
• Set up code to call post endpoint after detections.
• Set up template code
• Add Prisma ORM
• Set up database
• Set up deployment
• Add post endpoint
• Secure post endpoint
• Make detection component
• Get and list detections in UI