Conservation AI aims to harness machine learning for various conservation projects. The conservation AI site allows you to upload your media for automatic species detection using our AI models. Using the site you can upload single files or batch upload files to create an online libary of classified data all in one place. The conservation AI platform also supports real-time uploads from devices such as camera traps to enable the real-time detection and classification of animal species. Further information about the conservation AI platform can be found here.
For image classification the site supports .png .jpg .jpeg formats. Images should be limited to a resoultion no greater than 2000 x 2000 pixels to obatin the best results. For video processing the site supports .mp4 and .wmv formats. The site will extract and process one frame for every one second of video (1-minute video will equate to 60 individual .jpg images) which will be automatically analysed by the selected model. For the classification of audio the site supports .wav files. Audio files can be easily converted before uploading them to the conservation AI site.
Uploading Observations Through the Site
Media files can be uploaded one at a time or in batches. Batch processing takes around 10 minutes to process depending on the amount of files you upload. After processing the files will be saved to the relevent project in the results dashboard. You are required to have a conservation AI account to use the batch upload feature. We regulary release updates to improve the accuracy of our models as more data becomes available.
To upload media files to the conservation AI site click on the dashboard tab and select upload files. This process allows you to upload either single files or batch upload mutiple files.The same process can be repeated for audio files by clicking on upload single audio file.
Our models are trained to classify different species based on their geographical location so it is important to select the correct model to avoid misclassificaion. A list of models and associated species can be found here
Once processing is complete the results can be viewed in the project gallery. The maximum number of images per batch is 1000. The site utilises Graphics Processing Units (GPUs) to enable fast processing of uploaded content and is able to process 3,600 images per hour.
Uploading Observations Using Camera Traps Drones and Other Sensors
All of our AI models have their own distinct email address which can be used to email images directly to the model for automatic classification. Most camera traps support the transmission of images using the Simple Mail Transfer Protocal (SMTP) which can transmit images in near real-time. Once the images are processed the results will be available in the dashboard. Click here for a link to our models and their associated email address.
Frequently Asked Questions
If you are new to conservationai check out the Getting Started
guide to learn how to use some of the sites features. If you have a question about the site, please use the FAQ below. If you need help using the site or run into any issues please get in touch using the help
What is ConservationAI and how is it different from other platforms?
ConservationAI is a platform which uses machine learning to detect animals/object in images (both video and stills) and in audio files for a pre-determined set of species (see below).
How does the Artifical Intellegence work on the Conservation AI platform?
All of our models have been trained using a transfer learning approach which means we can generate new models using a significantly reduced dataset. We use different types of Deep Learning (DL) models for different applications. For image recognition we use deep convolutional neural networks which are trained and hosted using Googles TensorFlow framework. DL models get better at recognising different species by training on large and diverse datasets. We continually retrain our models as new images becomes available. For each model we append a number next to its name to indicate how many tags have been used for each class (species) to train the model. For example, the Sub Saharan 1k model has been trained on one thousand tags per class.
Can Conservation AI process images, videos and audio files?
Yes, it can process all of these.
What species can Conservation AI detect?
Please see our current models here. More species will be added as well as regions depending on our projects.
Are your models updated to improve accuracy?
Yes, as more images are tagged we will -re-train models to increase accuracy.
How often do you add new species?
This is very project dependent. Please contact us if you would like certain species added and can put some time into tagging these.
What is the maximum resolution of still images and video that I can upload for detection/classification?
The maximum is 2000 x 2000 pixels.
I have a large number of still images. Can I upload these at once in a batch?
Yes, you can upload a batch up to 1000 images.
Can I upload both stills and video?
Yes. For videos we automatically extract 1 frame per second for the analyses.
I have video images from a drone project and would like to detect/classify animals in it. How do I get started?
Please contact us. We can then discuss options to add a project, add images, so that you an start tagging and we can train a model for you. Our lower threshold for tags per species is 1000. More tags are better but we can start at 1000.
I would like to help with tagging. How do I get started?
Please see our instruction pdf here.
I have still images from a camera trap project. Can I upload these to create a project and tag?
At the moment please contact us and we can then discuss setting up a project for you in which we can upload your images so that you can start tagging.
The service seems free now, but will I have to pay if I have a large number of images?
The Conservation AI platform is currenlty free to use regardless of the number of imgaes you process.