Open-source imagery datasets are crucial for advancing machine learning in agriculture, as they enable training, validation, and comparison of modern models. This study captured Red Green Blue bands (RGB) and Normalized Difference Vegetation Index (NDVI) image dataset to characterize nitrogen deficiency symptoms in rice across growth stages (40–90 days after planting). A field experiment was conducted with four nitrogen levels: T0 (0 g N/plant - control), T1 (3.1 g N/plant - 50% of recommendation), T2 (6.2 g N/plant - 100% of recommendation), and T3 (9.3 g N/plant - 150% of recommendation) a complete randomized block method was used in the experimental field. Images were captured using a GoPro Hero 10 (RGB) and a Surveyor NDVI (Mapir Survey3) camera from multiple angles to ensure complete leaf coverage. The initial 12,000 images underwent automated and manual quality checks, resulting in a final dataset of 10,000 high-quality JPG images (5,000 RGB and 5,000 NDVI). Agronomic data (plant height, leaf area, tissue N, and grain yield) were analyzed using one-way ANOVA with Tukey’s HSD (P ≤ 0.05). Significant differences were observed among treatments, with the recommended nitrogen rate (T2) achieving optimal growth and yields comparable to T3, while T0 and T1 exhibited clear deficiency effects. The resulting dataset links visual symptoms with measured nitrogen status, providing a strong foundation for developing deep-learning models for classification of nitrogen deficiency detection in rice, especially for smallholder farming systems in Tanzania.