TensorFlight tutorial

Table of Contents

1 What is TensorFlight?

1.1 Core product

TensorFlight automatically detects and counts various types of objects from aerial imagery. We started with detecting trees, but we are also partially supporting:

  • Animals
  • Vehicles
  • Parking spaces

Please see some examples:

orchard_tutorial_example.jpg

facebook_tutorial_example.jpg

Our algorithms are based on machine learning. Even if algorithm initially does not do good job on your case, the more similar data like yours we see the more accurate we become.

Please help us adapting the product to your needs by completing a short survey.

1.2 In-depth classification

What's more, we are also supporting advanced analysis within a category. We are already distinguishing between healthy and dead trees. We are working on more in-depth classification algorithms, for example classifying tree type and age.

2 Upload drone imagery to DroneDeploy

TensorFlight works on drone imagery stored on the DroneDeploy app platform.

You can either upload data you already collected via drone to DroneDeploy, map a new area using DroneDeploy or upload processed orthomosaics from other platforms.

If you want to map a large area and you are not sure how to do it, please get in touch at support@tensorflight.com.

3 Install the TensorFlight app

apps_section.png

You can also open the app store directly at https://www.dronedeploy.com/app2/applications.

3.2 Pick TensorFlight from the list and install it

install_app.png

4 Request processing of the map

4.1 Select the map

Click the map you would like to process. select_map.png

4.2 Click "Analyze the map"

Expand the TensorFlight section by clicking the triangle button and click "Analyze the map".

analyze.png

4.3 Select an area

You may only interested in performing analysis on part of the map. In order to do so, select an option "Parts of the map" and pick annotations you are interested in.

You firstly need to create an Area annotation by clicking the "Area" button. We recommend adding a distinctive color or title for your new annotation.

Secondly, you need to pick annotations you are interested in and click the "Confirm" button. select_area_2.png

Don't worry, you will be able to change the selected area without restarting the whole processing.

4.4 Wait for processing

Wait until processing finishes:

processing.png

It should be finished within minutes for majority of maps, but can take a few hours for very big maps.

5 Analyze results

After processing is completed you should see a following view:

summary_view_2.png

5.1 Summary

Summary gives you a total count of objects in the selected area.

5.2 Detailed report

Detailed view lets you explore the view of the map with annotation for each object: detailed_view.png

5.3 Shapefile export

Shapefile lets you export results of the analysis to GIS software. Officially supported options are QGIS or ArcGIS.

For GIS experts - Shapefile consists of a layer named "TensorFlight" in EPSG 4326 projection with a polygon feature per object with a string field "type" representing object type - e.g. "Healthy tree" or "Parked vehicle" as seen in the examples in following sections.

5.4 Analyze Shapefile in QGIS

  1. Install QGIS.
  2. Unzip the exported zip and import the .shp file to QGIS like in the tutorial at https://www.youtube.com/watch?v=ehwoRkT-V7k.
  3. Select categorized layer style as described in http://www.qgistutorials.com/en/docs/basic_vector_styling.html. Optionally change the layer tranparency in the style section.
  4. Right click the layer in the "Layers Panel" section and tick "Show feature count".
  5. Optionally load your map as background in QGIS. Remember to use EPSG 4326 projection during the export.

You should results like this: qgis_export.png

5.5 Analyze Shapefile in ArcGIS online

In order to upload shapefile to ArcGis online:

  1. Add -> Add Layer from File -> Choose File -> Select zip file
  2. Select "Keep original features"
  3. Click "Import layer"
  4. Select the drawing style "Types"
  5. Optionally select the Basemap

You should see results like this: arcgis_export.png

6 Troubleshooting

6.1 Chrome cache

If we recently updated our app, your browser may still be using old version of our app. Problem should In order to force update to the latest version you may need to clear your browser cache. instructions for chrome, instructions for firefox, intructions for Microsoft Edge.

6.2 Detailed report does not load

Our current app loads all your imagery into the browser. Your computer may not have enough ram to process it. If you are dealing with very big maps, recommended way to analyze results is to export the shapefile and analyze results in QGIS or ArcGIS.

6.3 Results are not there yet after 24 hours.

We aim to process majority of requests within minutes and all reqests within 24 hours. We are sorry if you have been waiting longer. We are likely already working on solving your problem, but you may also contact our support, as decribed in the Ask for TensorFlight team's help section.

7 Ask for TensorFlight team's help

If something doesn't work as you expected please let us know at support@tensorflight.com.

The more data you provide, the better accuracy we provide back! If your map includes unusual objects or unusual situations it may not work imidiatelly, but each map you send us means that cases like yours will be handled better in the future. Don’t hesitate to send us any maps by using our app.

Please send us a DroneDeploy url of your map.