events.json: All multimodal event types
arguments.json: All arguments roles.
meed.json: Then whole dataset.
The images can be download at Baidu Netdisk : https://pan.baidu.com/s/1YC72EQVlk6xw9DkAgVoBzw code: zrxq
The format of the data:
{
"image_id": int,
"image_name": str,
"text": [str,str,str],
"event": str,
"trigger":{
"word":str,
"position":[{"start":int,"end":int},{"start":int,"end":int},{"start":int,"end":int}]
}
"arguments":[
"role":
{
"word":str,
"position":[
{"start":int,"end":int},
{"start":int,"end":int},
{"start":int,"end":int}
],
"bbox":[x1,y1,x2,y2]
}
]
}There are two example of dataset presented in the paper:
