If you cannot access the platform - not even the login page, check out status.picterra.ch to see if all our systems are operational. If that doesn’t help, reach out to support@picterra.ch

You can use any drone to capture your images: we support orthophotos coming from all the major photogrammetry software.

Ideally all the uploaded images should be georeferenced 8-bit RGBA geotiff (.tiff). This allows to exploit RGB color mosaics but also false color near-infrared images, thermal infrared images or digital surface models in the appropriate rasterized format (8-bit geotiff).

Other types of images are also supported (.png, .jpg, multispectral, single band, 16 or 32 bits) but will have to be normalized in order to be displayed and relevant information could be degraded or lost in the process. To best handle multispectral images (with more than 3 channels), refer to this article.

Yes, you can upload a raster with the surface or terrain information. However, you need to be careful with the normalization and ideally work with Height information in 8-bit format.

If you would like further information on how to best use your DSM or DTM to improve the results of your detectors you can check this article.

Picterra does not process raw drone images. The best way is to use a photogrammetry software to produce an orthophoto from your images ("stitching") and then upload the ortho.

If you upload individual drone images (e.g JPG), they will be considered as non georeferenced imagery. Check this article below to understand better the implications.

Picterra can support non-georeferenced images (e.g. .jpg, .png), however the platform is optimized for geo-referenced orthomosaics. As explained further below, if you upload image without georeferencing, the platform will arbitrarily geolocalize it at the (0,0) location, also called the "Null island" with an arbitrary resolution set to 10cm.

This means that when you run a detector on the image, you will get back latitude/longitude coordinates around this point zero. Our python client library exposes a function to convert the detector output back into image pixel coordinates. We will walk through its usage in this notebook.

Unfortunately you won't be able to access any of the Picterra features when you don't have an active subscription with Picterra. However, your data will be available and stored (for no more than 90 days after the termination of your subscription). If you need access to your data after termination, we strongly recommend downloading the information prior to your termination.

No, we have multiple users working with centimeter or sub-centimeter GSD.

The maximum image size is 50 GB (Enterprise plan) and 225'000 MP.

This can be either because your image is not georeferenced or because the georeferencing information is not recognized.

If your imagery is not already a georeferenced orthomosaic or image, you can georeference it manually using a GIS software and reference points from either another georeferenced raster or ground control points.

Here you will find documentation on how to georeference your image with QGIS using its georeferencing plugin.

If you want to proceed still without georeferencing, the platform will arbitrarily geolocalize it at the (0,0) location, also called the "Null island" with an arbitrary resolution set to 10cm. This allows then to work with this imagery on the platform "as if" it was localized. Beware that if then you download the image from the platform, you will get the one geolocalized by ourselves arbitrarily at (0,0).

If the image is using an uncommon spatial reference system, it is possible Picterra doesn't recognize it.

In this case, one way to fix the problem is to load the image in QGIS and save it again by changing the CRS to EPSG:4326 / WGS84 and then upload that new image to Picterra.

The MegaPixels (MP) of an image is the total number of pixels constituting the image, per Million pixels. So 1 MP is equivalent to saying there is 1Million pixels in this image.

This is important since more MP means longer detection time (the model needs to see more pixels).

You can at anytime on Picterra platform check the number of MP of an image:

MegaPixels calculator

This heavily depends on the use case. For production usage requiring high accuracy, we usually see users using few hundreds to low thousands of annotations.

That being said, one of the powerful feature of Picterra is that you can start small (20-50) and then iterate from there as you add more annotations to improve your detector.

Indeed the key to creating good training, accuracy and testing areas is representativeness. The regions covered by your areas should be representative of the areas that you’ll be trying to run detections over as a whole. You should not have that many redundant training areas (over too many similar regions for example). Rather focus on variety and quality of your areas and annotations, not quantity.

For segmentation mode detector, the accuracy to use it the Foreground IOU (or intersection over union, also known as the Jaccard index.) This is a measure of how well your detected regions overlap with the regions you outlined in your Accuracy Areas. This is a “per pixel” score, meaning it is not suited for matching detected and annotated objects. If you are trying to detect individual objects you should be in count mode, not segmentation mode.

For count mode detectors, the accuracy scores to use are the recall and precision. The recall is the percentage of your outlined objects in your accuracy area that were correctly matched by a detection. The precision is the percentage of your detections that matched an outline. Combined, these can be used to easily compute a single score called the F-score.

Sign up for Picterra University for a more in depth understanding on detector accuracy as well as many of the other advanced features of the platform.

If you have annotations on an existing detector and want to transfer them to another detector, you can do this by first downloading the annotations from your first detector, split them up into multiple files by category in a GIS software, and then import them to your second detector.

Picterra has developed an ArcGIS Add-in (plugin) allowing to directly push data, run detectors and retrieve output results from within ArcGIS.

More information on the Add-in and where to download it in this section.