Using deep learning methods, it is best to use Pytorch to implement a visual odometer network, with some Github and some papers for reference

I am looking for a skilled Pytorch developer to implement a visual odometer network using deep learning methods. The ideal candidate should have experience working with Github and referencing papers. The network works like DeepVO, input the images before and after frames, and predict its 6Dof pose. However, the accuracy of DeepVO is too low. If possible, I hope to add a segmentation network to the decoder section. Of course, implementing a visual odometer network is the most important. Accuracy: The desired accuracy level for the visual odometer network is high.(the error should not be too high) Architecture: There is no preferred deep learning architecture, so the candidate can use their expertise to choose the best one for the project .You can use pure CNN, or combine RNN/LSTM with CNN, etc. Deadline: The project has a normal deadline of 1-3 weeks. I am not proficient in using Freelancer. If you are interested, please contact me.