Elementary Concept Reasoning (ECR) is a novel dataset focusing on object-centric visual concept learning. It contains RGB images (64×64×3) of 2D geometric objects on a constant colored background. Objects can vary in shape (circle, triangle, square, and pentagon),size (large and small), and color (red, green, blue, yellow). Uniform jitter was added to each color, resulting in various color shades. Each image contains a single object which is fixed to the center of the image. Furthermore, ECR contains image pairs following the match pairing setup.