Multimodal Visual Instruction Dataset for Intelligent Transmission Line Inspection: Single-Turn Fine-Grained Description Edition

The response is written according to the actual visible content and describes the inspection environment, equipment type and structure, anomaly location, appearance characteristics, defect condition, potential operational risks, and recommended maintenance actions.The dataset follows a conversational instruction format suitable for fine-tuning multimodal large language models such as Qwen-VL. Images are referenced using relative paths to facilitate data migration, model training, and experiment reproduction.

Before release, the dataset was checked for image-annotation pairing, file readability, duplicate records, train-validation leakage, category validity, semantic coverage, conversation structure, and path portability. The responses do not contain object-detection procedures, region coordinates, or bounding-box descriptions. The dataset can support research on transmission-line image captioning, visual question answering, defect recognition, condition assessment, risk analysis, multimodal instruction tuning, and intelligent inspection report generation.