Available models

The algorithms in zamba are designed to detect and classify animals that appear in camera trap images and videos. This section is organized first by the type of data you have (image vs. video) and then by the task you want to perform.

Which model should I use?

Data type	Task	Models
Images	Species classification	lila.science (default), speciesnet
Videos	Species classification	time_distributed (default), slowfast, european
Videos	Blank detection	blank_nonblank
Videos	Depth estimation	depth
Videos	Pose estimation	DensePose

All of the species classification models support training, fine-tuning, and inference. For fine-tuning video models, we recommend starting from time_distributed; for image models, we recommend starting from lila.science.

All of the image and video species classification models share a common first step: an efficient object detector (MegaDetector / MegadetectorLite) is used to find the animals in a frame before classification.

Model summaries

Image models

Model	Geography	Relative strengths	Architecture	Number of training images
lila.science	Global based on datasets from lila.science	Good base model for common global species. Default image model.	ConvNextV2 backbone	15 million annotations from 7 million images
speciesnet	Global	Very large taxonomy (2,000+ classes); strong alternative starting point for fine-tuning on global species.	EfficientNetV2-M backbone	Google's SpeciesNet training set (tens of millions of images)

See the image species classification page for details.

Video models

Model	Geography	Relative strengths	Architecture	Number of training videos
blank_nonblank	Central Africa, West Africa, and Western Europe	Just blank detection, without species classification	Image-based TimeDistributedEfficientNet	~263,000
time_distributed	Central and West Africa	Recommended species classification model for jungle ecologies	Image-based TimeDistributedEfficientNet	~250,000
slowfast	Central and West Africa	Potentially better than time_distributed at small species detection	Video-native SlowFast	~15,000
european	Western Europe	Trained on non-jungle ecologies	Finetuned time_distributedmodel	~13,000

See the video species classification and blank detection pages for details.

The models trained on the largest datasets took a couple weeks to train on a single GPU machine. Some models will be updated in the future, and you can always check the changelog to see if there have been updates.

User contributed models

We encourage people to share their custom models trained with Zamba. If you train a model and want to make it available, please add it to the Model Zoo Wiki for others to be able to use!

To use one of these models, download the weights file and the configuration file from the Model Zoo Wiki. You'll need to create a configuration yaml to use that at least contains the same video_loader_config from the configuration yaml you downloaded. Then you can run the model with:

$ zamba predict --checkpoint downloaded_weights.ckpt --config predict_config.yaml