Taiwan Train and Validation Data for ICDAR'25 MapText Competition

Important: v1.1 Fix two invalid polygons in tw25_train.json (indexed at 1155 and 1156). Dataset of 2Kx2K image tiles cropped from Taiwan Historical Maps System for the ICDAR'25 Competition on Historical Map Text Detection, Recognition, and Linking. This collection contains topographic maps related to Taiwan, covering detailed records of the terrain, topography, and geographic names (in traditional Chinese). Annotations and images follow the format described at the competition website and can be evaluated using the official evaluation repository script. Please note the we also provide an extra synthetic dataset in traditional Chinese for training, which is released under the record: "SynthMap+ (Traditional Chinese) Synthetic Train Data for ICDAR'25 MapText Competition" (10.5281/zenodo.14502179). Train Validation Annotations tw25_train.json tw25_val.json Images train.zip val.zip Files tw25/train/*.jpg tw25/val/*.jpg Tiles 1,478 166 Map Sheets 169 30 Words 13,153 5,007 Label Groups - - Illegible Words 2,719 794 Truncated Words 2,544 578 Valid Words 10,434 4,213 Important: All the valid characters shall fall in the following unicode range (otherwise ``illegible'' of the text instance is set to True): 0x3400-0x4DBF, 0x4E00-0x9FFF, 0xF900-0xFAFF, 0x20000-0x2A6DF, 0x2A700-0x2EBEF