Limited/No detections available in older imagery

Hello,

Is there a limited amount of detections available for older imagery?

In particular, I am trying to fetch “nature–vegetation” and “nature–terrain” detections in images from 2017, and there are none at all. I have checked around 45k of them (through a script that talks to the graph api).

Worth noting that I am able to get tons of such detections from imagery in 2021 and 2022 (see example image attached).

Thanks a lot!
George

For example, here is a detections response for an example batch of 20 images.

>>> pprint(img_features)
{'108543031310984': {'height': 3000, 'id': '108543031310984', 'width': 4000},
 '203158914801104': {'height': 3000, 'id': '203158914801104', 'width': 4000},
 '203930774712867': {'height': 3000, 'id': '203930774712867', 'width': 4000},
 '238298038049209': {'height': 3000, 'id': '238298038049209', 'width': 4000},
 '240691767832543': {'height': 3000, 'id': '240691767832543', 'width': 4000},
 '280356203813802': {'height': 3000, 'id': '280356203813802', 'width': 4000},
 '317311680013031': {'height': 3000, 'id': '317311680013031', 'width': 4000},
 '4087492507961146': {'height': 3000, 'id': '4087492507961146', 'width': 4000},
 '4539489152727750': {'detections': {'data': [{'geometry': 'GjgKBm1weS1vchIYEgIAABgDIhAJ4C/aEhqyBAAAoASxBAAPGgR0eXBlIgkKB3BvbHlnb24ogCB4AQ==',
                                               'id': '4570407579635907',
                                               'value': 'warning--road-narrows--g1'}]},
                      'height': 3000,
                      'id': '4539489152727750',
                      'width': 4000},
 '455695235499217': {'height': 3000, 'id': '455695235499217', 'width': 4000},
 '457662825313831': {'height': 3000, 'id': '457662825313831', 'width': 4000},
 '463323258262180': {'height': 3000, 'id': '463323258262180', 'width': 4000},
 '473988820482591': {'height': 3000, 'id': '473988820482591', 'width': 4000},
 '478570490029705': {'height': 3000, 'id': '478570490029705', 'width': 4000},
 '479174056741227': {'height': 3000, 'id': '479174056741227', 'width': 4000},
 '485698512854670': {'height': 3000, 'id': '485698512854670', 'width': 4000},
 '489420462491614': {'height': 3000, 'id': '489420462491614', 'width': 4000},
 '491433148583383': {'height': 3000, 'id': '491433148583383', 'width': 4000},
 '516777993033710': {'height': 3000, 'id': '516777993033710', 'width': 4000},
 '527903208225456': {'height': 3000, 'id': '527903208225456', 'width': 4000}}

As you can see, for 20 images of 2017 (the image ids are the leftmost number), I get only a single detection (warning–road-narrows–g1) in just one of them.

So this doesn’t look right!

How did you downloaded the semantic segmentation data? I couldn’t manage to achieve this…

Hello kvestena!

I am in the process of developing a library (in this GitHub repo: GitHub - gchoumos/mapillary: Tools to interact with the Mapillary API (v4) ) to do a few things with the new (v4) Mapillary API. Up to now the developed functionalities include

  1. Get image sequences of an organization or a user
  2. Get images (image ids)
  3. Get image metadata
  4. Download images
  5. Get detections of images
  6. Visualize the detections on top of the images

So you can have a look at this library and see if it fits your needs. It will be updated further! Bear in mind that the current documentation doesn’t cover the detections (the code is there of course, but I just didn’t find the time to explain it in the README yet).

So, to be more specific to your question, I downloaded the detections (the geometries and labels) through the Mapillary API. Then, to display them on top of the images, I did the necessary transformations as described in the Mapillary documentation.

Cheers,
George

1 Like

Oh, really thank you, @gchoumos !!!

I just want to know the percent of image that is filled with vegetation (it can include terrain as well),

I was trying to find some pretrained model, but isn’t necessary, as there’s the Mappilary detections, but since the transition to API v4 they seem to have been vanished…

I want to use it as a simplified metric of presence of vegetation in urban enviroments, likewise in this map (map legend only in portuguese): snav/current_version.pdf at master · kauevestena/snav · GitHub !!

Your post was the first hope to retrieve the vegetation segmentation!!