- Task 1 - Character Recognition - Method: MixNet based on multiple classic CNN
- Method info
- Samples list
- Per sample details
method: MixNet based on multiple classic CNN2019-04-30
Authors: 周圣杰(Shengjie Zhou)、黄富伟(Fuwei Huang)、文荟俨(Huiyan Wen)、李贤彪(Xianbiao Li)、周逸豪(Yihao Zhou)、苏统华(Tonghua Su)
Description: Organization:Harbin Institute of Technology
Contact information:15179126343 (Fuwei Huang)
Description:
For training set, we've made an ample expansion. Specifically, we rotated images where its angale between ten degrees. In addition, we adjusted the pictures' brightness and constrast randomly. Finally, we got a dataset which is 40 times bigger than the original dataset. With regards to the network, we developed net which mixed ResNet, AlexNet, etc. Our team did a lot of fine-tuning on for each classical CNN, in order that they can work together. At last, we did a vote based on ResNet's and AlexNet's result to get the result.