Data Labeling with CVAT

Starting CVAT in Windows

Assumes you’ve already completed the installation instructions.

1. Open Docker App to start the Docker engine

2. Open Up Ubuntu App command prompt

3. In the Ubuntu prompt

   1. cd cvat

   2. Without SAM

      1. docker compose up -d

   3. With SAM

      1. docker compose -f docker-compose.yml -f components/serverless/docker-compose.serverless.yml up -d

      2. cd serverless && ./deploy_cpu.sh pytorch/facebookresearch/sam/nuclio/

4. Open Google Chrome

5. in the address bar, go to localhost:8080

6. This should bring up the CVAT log in screen. Log in with your username and password that you created during installation.

7. Go to your project and start labeling (this assume you’ve already created a project and imported your dataset)

Creating a Project

1. Download data from the shared google drive folder that @Matthew Kutugata shares with you

   1. Download the data from your “unlabeled” folder

2. In the CVAT app, go to the “projects” tab, create a new project by clicking the plus button

3. Copy the downloaded zip file name and use it as the project name

   1. For example, “very_large_hairy_vetch_MD_2022-10-27_annotations_camvid”

4. Press “submit and continue” to make the project

Importing Data

1. Select the newly created project

2. Choose the actions and select “import dataset”

3. Select “CamVid 1.0” for the import format

4. Click or drag the downloaded zip file

   1. the zip file name should now be under the “drag here” box

5. Press “ok” to start the import

   1. An initial progress notification should pop up on the screen

   2. Then another “Annotation import finished” notification should pop-up. This may take some time (~1-2 minutes).

   3. You should see the dataset in the project

Exporting Data

1. Make sure you save your project by pressing the “Save” icon

2. press the Menu (3 lines) icon

3. export job dataset

4. Choose CamVid 1.0 format

5. Give the export a proper custom name. I need

   1. number of images

6. Press ok

7. The download request will be in the “request” tab. It may take a little bit of time to prepare the download

8. Once it’s finished, press the 3 dots to “download”

9. Once downloaded onto your local computer, place the completed job zip folder in the appropriate google drive folder.

Labeling

Image prep

1. Remove the background class overlay for each image

2. Remove all background classes when starting out

Polygon Tool

1. Select the polygon tool from the tool bar (pentagon)

2. Select your target class, not “background”.

   1. You should only have 1 target class and a “background” class. Choose the target class

3. Outline small part of the plant that are missing

   1. Once you select the polygon tool and the class, choose shape

   2. The polygon tool should now be active on your screen (cross symbol)

   3. Use the scroll wheel to pan across the image

   4. Left-click to select points

   5. hold shift to automatically select point with every movement of your mouse

   6. Press n again to close the polygon

   7. Press n again to reactivate the polygon tool

   8. Save often

4. Overlapping other section is ok

5. Use the “bitmap” to check your work

6.  

Uploading Files

1. Log into Google Account

2. Go to Private Google Drive for CVAT

   1. It’s a good idea to have this link bookmarked

3. Open File named ‘Labeled’

4. In the upper lefthand corner of Google Drive, use “+ New” to upload a folder

5. Drag and drop the folder into Drive

6. Name the Folder, SPECIES, SITE, DATE of collection

Shutdown Docker

docker compose -f docker-compose.yml -f components/serverless/docker-compose.serverless.yml down

and close docker

Keyboard shortcuts:

Installation:

https://docs.cvat.ai/docs/administration/basics/installation/

Setting Up SAM for CVAT

The Facebook Segment Anything Model is now available in the self-hosted version of CVAT | CVAT Blog

1. https://docs.cvat.ai/docs/administration/advanced/installation_automatic_annotation/

2.