FishMarket BD: A Dataset of Local Market Freshwater Fish Species in Bangladesh for Deep Learning and Computer Vision

Dataset Overview: This dataset comprises 1,581 images of ten popular freshwater fish species commonly found in local Bangladeshi markets. The images were captured under ambient retail market lighting conditions across varying angles, orientations, and background surfaces, and are organized into ten species-based class folders. Class DistributionSilver Carp: 341 Catla: 321 Mrigal: 302 Baim (Zig-zag Eel): 124 Tengra (Mystus Catfish): 114 Telapiya (Tilapia): 100 Pabda (Pabdah Catfish): 100 Roi (Rohu / Rui): 70 Poti (Swamp Barb / Puti): 62 Bowal (Wallago Catfish): 47 Data ProcessingOriginal raw market images were standardized to a 3-channel RGB color space, resized to a uniform $224 \times 224$ pixels matrix, and normalized by scaling pixel intensity values from $[0, 255]$ down to $[0.0, 1.0]$. Operational Python pipeline scripts (preprocess_pipeline.py, augment_generator.py) are provided in the Preprocessing_Code/ folder for complete operational reproducibility. Folder Structure & NamingThe dataset is structured into ten species directories within both raw (Raw_Dataset/) and processed (Processed_Dataset/) subfolders. Individual image files are assigned standardized unique identifiers (e.g., FMBD_0001.jpg) and mapped item-by-item with local, English, and scientific taxonomy names in metadata.csv within the Metadata/ folder. PurposeThis dataset supports the development and evaluation of lightweight automated fish species recognition systems using computer vision and deep learning. It is structured for benchmarking resource-efficient transfer learning models (e.g., MobileNetV2), validating Explainable AI (XAI / Grad-CAM) feature heatmaps, and powering real-world applications in smart aquaculture, fisheries management, food authenticity, and digital market sorting.