LaSOT is a high-quality benchmark for large-scale single object tracking, consisting of 1,400 sequences with over 3.5 million frames. Each frame is manually annotated with a bounding box, making LaSOT the largest densely annotated tracking benchmark. The average video length is over 2,500 frames, and each sequence includes various challenges such as target disappearance and re-occurrence. LaSOT provides high-quality dense annotations and additional natural language specifications to encourage the exploration of linguistic features for tracking. It includes 70 categories with 20 sequences each, and evaluates 35 tracking algorithms, showing there is still room for improvement. LaSOT addresses issues such as small-scale datasets, lack of high-quality annotations, short-term tracking, and category bias. It provides a large-scale, long-term tracking benchmark with comprehensive labeling and attributes for performance analysis. LaSOT is evaluated using two protocols, with results showing that deep trackers like MDNet and VITAL perform well, while others like SiamFC are efficient. LaSOT aims to provide a dedicated platform for training and assessing tracking algorithms, and to bridge the gap between deep trackers and large-scale datasets.LaSOT is a high-quality benchmark for large-scale single object tracking, consisting of 1,400 sequences with over 3.5 million frames. Each frame is manually annotated with a bounding box, making LaSOT the largest densely annotated tracking benchmark. The average video length is over 2,500 frames, and each sequence includes various challenges such as target disappearance and re-occurrence. LaSOT provides high-quality dense annotations and additional natural language specifications to encourage the exploration of linguistic features for tracking. It includes 70 categories with 20 sequences each, and evaluates 35 tracking algorithms, showing there is still room for improvement. LaSOT addresses issues such as small-scale datasets, lack of high-quality annotations, short-term tracking, and category bias. It provides a large-scale, long-term tracking benchmark with comprehensive labeling and attributes for performance analysis. LaSOT is evaluated using two protocols, with results showing that deep trackers like MDNet and VITAL perform well, while others like SiamFC are efficient. LaSOT aims to provide a dedicated platform for training and assessing tracking algorithms, and to bridge the gap between deep trackers and large-scale datasets.