redbioma/cemede-redbioma-ct

Authors

Salinas-Acosta, Adolfo; López-Venegas, María; Biarreta-Portillo, María; Morataya-Sandoval, Pamela; Víquez-Mora, Emilia; Mora-Cross, María; Gómez-Solís, William and Bautista-Solís, Pavel

Description

The cemede-redbioma-ct dataset contains 11,674 camera trap images (training 8088; validation 1801; and test 1785) of wildlife from Costa Rica, collected between 2022 and 2024 by the Mesoamerican Center for Sustainable Development of the Dry Tropics (CEMEDE, UNA) and processed by the Data Science Network for the Conservation of Mesoamerican Biodiversity (Redbioma) project. It covers 26 species of mammals, birds, and reptiles, with images extracted from more than 9,000 videos using MegaDetector. The dataset includes challenging conditions such as motion blur, rain, low light, and partial animal appearances, making it a valuable resource for developing and benchmarking computer vision methods in tropical biodiversity monitoring. Images are split into training, validation, and test sets at the video level to prevent data leakage. This dataset supports research on wildlife classification, class imbalance, rare species detection, and conservation applications.

The videos used to extract the images were recorded at the Nicoya campus of the National University of Costa Rica, where the SCALLUNA rainwater harvesting project was developed as a system for sustainable water management. Camera traps were deployed along the El Cornizuelo Trail, which provides year-round access for wildlife, particularly during the dry season. The site is a 10.6 ha fragment of dry tropical forest adjacent to the La Cruz Protected Zone, part of the Yaguarundi Forests Biological Corridor in the Tempisque Conservation Area. This corridor connects the counties of Nicoya, Santa Cruz, and Hojancha, and emphasizes water management, forest conservation, territorial connectivity, and ecological planning. Camera traps (Bushnell HD and Browning Strike Force models) operated from 2022 to 2024 generated the dataset.

The dataset preparation began by extracting frames from the input videos using MegaDetector. To reduce redundancy and avoid selecting nearly identical frames, only three frames containing animals were retained per video, skipping ten frames between each selection. Each image was then passed again through MegaDetector to obtain bounding boxes, which were required to train both the classification models and the object detection model. Keep in mind that the dataset contains the common names in Spanish for the species found in the image. The corresponding scientific name can be found in the summary of image counts by species, where file_name is the name found in the dataset.

The original unprocessed version of this dataset, containing the raw camera trap images (before being processed and cropped with the MegaDetector detections), is available at https://huggingface.co/datasets/redbioma/cemede-redbioma-ct-original. This version provides the original data from which the current dataset was derived.

Summary of image counts by species

Scientific Name	Test	Train	Val	Total	file_name
Nasua narica	149	700	151	1000	Pizote
Columba livia	150	699	151	1000	Paloma
Crypturellus cinnamomeus	149	700	151	1000	Tinamu
Agouti paca	148	700	152	1000	Tepezcuintle
Odocoileus virginianus	150	699	151	1000	Venado
Ortalis cinereiceps	144	678	147	969	Chachalaca
Didelphis marsupialis	141	663	144	948	ZorroPelon
Zenaida asiatica	111	516	114	741	PalomaAliblanca
Nyctridromus albicollis	81	375	82	538	Cuyeo
Turdus grayi	66	309	69	444	Yiguirro
Piaya cayana	38	186	42	266	CucoArdilla
Sciurus variegatoides	36	168	46	250	Ardilla
Iguana iguana	37	168	39	244	Garrobo
Urocyon cinereoargenteus	34	174	33	241	ZorroGris
Coragyps atratus	36	149	53	238	Zopilote
Aramides cajaneus	35	156	41	232	Rascon
Canis latrans	64	156	9	229	Coyote
Falco peregrinus	33	152	37	222	HalconPeregrino
Conepatus semistriatus	35	126	37	198	Mofeta
Crotophaga sulcirostris	27	110	28	165	Tijo
Icterus auricapillus	30	114	20	164	Oropendola
Rhinella marina	18	91	26	135	Sapo
Megascops cooperi	17	78	25	120	Buho
Icterus galbula	18	75	18	111	OropendolaDeBaltimore
Myiarchus nuttingi	19	76	15	110	Copeton
Tyrannus melancholicus	19	70	20	109	TiranoMelancolico

Citation

This dataset can be used for investigation purposes including both the appropriate citations:

@misc{cemede_redbioma_ct,
 author       = {Salinas-Acosta, Adolfo and López-Venegas, María and Biarreta-Portillo, María and Morataya-Sandoval, Pamela and Víquez-Mora, Emilia and Mora-Cross, María and Gómez-Solís, William and Bautista-Solís, Pavel},
  title         = {redbioma/cemede-redbioma-ct},
  year       = {2025},
  month    = aug,
  note       = {Dataset published in Hugging Face, 31 August 2025},
  url          = {https://huggingface.co/datasets/redbioma/cemede-redbioma-ct}
}

PENDING: Paper

License

This dataset is released under the Creative Commons Attribution 4.0 International (CC BY 4.0) license. You are free to share and adapt it for any purpose, provided that appropriate credit is given, a link to the license is included, and any changes are indicated. Full license available at https://creativecommons.org/licenses/by/4.0/