ConstantHao/EDU-CHEMC_MM23
收藏资源简介:
We now released the EDU-CHEMC dataset, which was initially proposed in Paper "Handwritten Chemical Structure Image to structure-Specific Markup Using Random Conditional Guided Decoder". All images are in folder EDU-CHEMC. We provide an annotation json file for each image with same filename. In each json, there are three keys: * chemfig: the origin chemfg string annotated by humans, which can be rendered with textlive when use chemfg package * ssml_sd: the training target of SSSL-SD, where modeling units can be seperated by space. One can directly feed modeling units into encoder-decoder models. You can also obtain the graph format by simply parsing. * ssml_rcgd: the training targets of SSSL-RCGD. Since there are multiple SSSL-RCGD targets for a training image, we randomly sampled some of them so the value corresponding to this key is an array. * We now explain the format of SSSL-RCGD. Each training target of SSSL-RCGD is an array sorted by time step, where each element is a three-tuple (text, reconnection marks, condition input) * text: string, the base modeling unit * condition input: int, the index (start from 0) of the candidate BAU(Branch Angle Unit) to be fed into decoder at current time step * reconnection marks: an array, where each element is a two-tuple (reconnection index, bond type) * reconnection index: the index the candidate BAU which forms a reconnection with element of current time step * bond type: string, the bond type of reconnection * ssml_normed: the submission format of ICDAR2024 CROCS Competition, see https://crocs-ifly-ustc.github.io/crocs/data.html We also provide EDU-CHEMC.vocab which contains all modeling units.




