Detect Images: Extra topics

What species can be detected?

We currently detect 70 Southern African species, including humans and vehicles. Additionally we flag empty images as “No animal”.

Some species are grouped, for example we currently detect both Black and Blue Wildebeest as simply “Wildebeest”.

See also Species list.

See also Additional labels.


We are actively adding more species as training data for these species becomes available.

You can assist by requesting new species, and supplying sample images.

Over time we will refine the training within the grouped species, and for example detect Black and Blue Wildebeest separately.

Species list

The following species have been trained and can be returned by the model.

Some species are grouped. For example, the model will currently only return “Hare”, but has been trained with images of Scrub Hare, Savannah Hare, Cape Hare, and all of these labels are available to you if you would like to refine your data further.

Numbered items in the list can be returned by the model; bulleted items can be refined by you.

  1. Aardvark

  2. Aardwolf

  3. Baboon

  4. Bird (Other)
    • Bird (of Prey)

    • Bird (Other)

    • Crane (Blue)

    • Francolin - NEW

    • Spurfowl - NEW

  5. Blesbok

  6. Buffalo

  7. Bushbuck

  8. Bushpig

  9. Bustard
    • Bustard (Kori)

    • Bustard (Ludwig’s)

    • Bustard (White Bellied)

  10. Caracal

  11. Cat (African Wild)

  12. Cheetah

  13. Civet

  14. Dikdik

  15. Domestic Animal (Cattle)

  16. Domestic Animal (Dog)

  17. Duiker
    • Duiker (Common Grey)

    • Duiker Natal Red)

  18. Eland

  19. Elephant

  20. Fallow Deer - NEW

  21. Fox
    • Fox (Bat-eared)

    • Fox (Cape)

  22. Gazelle (Grant’s)

  23. Gazelle (Thomson’s)

  24. Gemsbok

  25. Genet
    • Genet (Common Small Spotted)

    • Genet (Large Spotted) - NEW

  26. Giraffe

  27. Grysbok - NEW

  28. Guineafowl

  29. Hare
    • Hare (Savannah)

    • Hare (Scrub)

    • Hare (Cape)

  30. Hartebeest
    • Hartesbeest (Red)

  31. Hippopotamus

  32. Honeybadger

  33. Human

  34. Hyena (Brown)

  35. Hyena (Spotted)

  36. Hyena (Striped)

  37. Hyrax (Rock) - NEW

  38. Impala

  39. Jackal
    • Jackal (Black-backed)

    • Jackal (Side-striped)

  40. Klipspringer

  41. Kudu

  42. Leopard

  43. Lion

  44. Mongoose
    • Meerkat

    • Mongoose (Egyptian Large Grey)

    • Mongoose (Slender)

    • Mongoose (Small Cape Grey)

    • Mongoose (Yellow)

    • Mongoose (White Tailed)

  45. Mongoose (Water)

  46. Monkey (Vervet)

  47. Nyala

  48. Ostrich

  49. Porcupine

  50. Rabbit
    • Rabbit (Red Rock)

    • Rabbit (Riverine)

  51. Reedbuck
    • Reedbuck (Mountain)

  52. Reptile Amphibian

  53. Rhebok (Grey)

  54. Rhinoceros
    • Rhinoceros (Black)

    • Rhinoceros (White)

  55. Rodent
    • Rodent

    • Squirrel (Ground)

  56. Secretary Bird

  57. Serval

  58. Springbok

  59. Springhare

  60. Steenbok

  61. Striped Polecat

  62. Topi

  63. Tortoise

  64. Vehicle

  65. Vulture

  66. Warthog

  67. Waterbuck

  68. Wild Dog - NEW

  69. Wildebeest
    • Wildebeest (Black)

    • Wildebeest (Blue)

  70. Zebra
    • Zebra (Burchell’s)

    • Zebra (Mountain)

“No animal” is also returned by the model for empty images.

See also Additional labels.

Additional labels

The following species have not yet been trained, and so will not be returned by the model, but the labels are available to you in WildID for labelling your images:

  1. Bat

  2. Bontebok - NEW

  3. Crocodile

  4. Domestic Animal (Other)

  5. Domestic Animal (Sheep/Goat) - NEW

  6. Error

  7. Fire

  8. Insect Spider

  9. Monkey (Colobus)

  10. Monkey (Sykes)

  11. Oribi

  12. Other

  13. Otter

  14. Pangolin

  15. Roan

  16. Sable

  17. Seal (Grey) - NEW

  18. Seal (Harbour) - NEW

  19. Unknown

  20. Weasel (African Snake)

  21. Weasel (White Nape)

  22. Vehicle (Bike)

How do you know if your image has been detected or not?

Before detecting an image, you can see the image in your Manage Images list, in your CSV export, and view the image itself and its details in the Image Details screen, but it will have no labels listed or boxes displayed.

You will be able to search for the image by camera or filename, but not by species, as WildID doesn’t yet know what species are present in the image. The image will thus not be included in your species counts in the Dashboard.

Example WildID screen showing Card view with images that have and have not yet been detected.

Can you re-detect images?

Right now you cannot re-detect images in WildID through the website.

We are able to do so in the backend, so if you need to do this for a good reason, please contact us to discuss.


We will be adding the ability to re-submit selected images for re-detection.

For example, the current model may not find an animal that does exist in one of your images, but a future model, better trained, may be able to pick it up. We’ll aim to let you resubmit especially “No animal” images, and possibly those images you have marked as incorrect.

We will keep a history of the results from each detection, so that it is possible to see how the model matures over time.

Detection queue priorities and speed

We are currently processing camera trap images at a rate of about 5 seconds per image.

We prioritise images for detection in the following way:

  • Single detect images submitted through the WildID website are submitted with priority 1.

  • Bulk detect images submitted through the WildID website are submitted with priority 2.

  • Huge numbers of images uploaded, for example when an organization comes on board and has a large image backlog, are submitted with priority 3.

At any one time, the detection model will look for the next image to process, and select submitted images in order of priority. This means that if an Organisation 1 has submitted TB worth of data, and Organization 2 has submitted 500 images through bulk detect in the website, and you in Organisation 3 are doing a demo to your colleagues and press the Single Detect, your Single detect will get processed first, and when the detection engine has some spare time it will focus on the bulk detect and then the backlog images.

If you need to process a batch of images more urgently than normal for a good reason, please speak to us about increasing the priority of your image batch.


We monitor the processing speeds, and as pressure on the queues gets higher (more users with more images) and the time from submitting your images for detection to the time when all your images are processed gets longer, we will add additional queues and servers based on demand.

Data returned and stored when you detect an image

When you submit an image for detection, WildID submits the image via API to our machine learning model, which evaluates the image, and returns predictions. For each object that the model finds and recognizes in the image, it returns:

  • A localization or bounding box – a box drawn around the animal

  • A species name

  • A prediction score out of 1 (0.6312 = 63%). This indicates how sure the model feels about the identification.

  • We do not return predictions with a score below 50%.

We store the predictions from the model in our database, as well as making a copy of them as the current object labels.

If you change the object labels, or delete any of the boxes and labels, you change the current object labels, but not the original predictions.


We will add a history of label changes to the image details screen, so that you will be able to see what the original predictions were, which version of the model was used for the detection, what the labels have been changed to and by whom, possibly with the ability to reset to the original labels or to any changeset.

Can WildID use my images to retrain a model?

Yes, we would like to do so. The more correctly labelled image data we have to train with, the more accurate the model will get. There is also a big advantage to you to have images from your camera traps (sites) used in training, as the model learns what species are likely to occur at various sites, and becomes more accurate at that site. There are also quirks with each camera trap model which mean that as we use examples from more models the accuracy also improves.

There is no extra cost to you if we use some of your images for training, and no special procedure or data preparation that you need to do. Continue uploading and verifying your images in WildID, and we will be able to access the data from there, with no interruption to you. No third party will be able to see your data or access it.

Can WildID train a model specifically for my use?

Yes, we can. If you have a very specific application, for example you would like to train a model only to recognise certain species outside Southern Africa, or to do recognition on aerial photographs, visitor counts and so on, or simply to be optimised for your sites only, we can train a model using only your data, and for use only by you.

You would need to supply the training data and species list that you would like trained.

This would incur additional cost for you. Please contact us to discuss possibilities.

Using WildID to generate training data for your own machine learning model

If you have camera trap image data that has no annotations yet (bounding boxes or object classifications), you can use WildID to help prepare your data for training a model.

  • Import your images into WildID.

  • WildID will perform initial detections and identifications, and mark empty images for you.

  • You can edit the identified labels if incorrect.

  • Even if your species are way different to those that WildID recognises (for example you are from a different geographic region), talk to us about remapping your results in bulk.

  • if you have species labels for your images but no bounding boxes, we can script the species labels you supply for each image to all bounding boxes detected by WildID.

  • View and select your training images easily in WildID.

  • WildID can export selected images and the related XML data for TensorFlow records required for model training.

Email us at if you would like to discuss possibilities for data preparation.