search

aofrancani / TSformer-VO

Implementation of the paper "Transformer-based model for monocular visual odometry: a video understanding approach".
https://arxiv.org/abs/2305.06121
MIT License
71 stars 11 forks source link

result #5

Closed ztx991028 closed 1 year ago

aofrancani commented 1 year ago

It is hard to tell from just one image. Maybe you can tell me the hyperparameters you used in training and how many epochs the model was trained on. Furthermore, there might be a natural variance in the weights during training, so be sure you are testing the best model according to the validation data.

ztx991028 commented 1 year ago

The parameters have not changed, I changed the batch_size to 2, epoch to 200, and you give 4 and 100.

ztx991028 commented 1 year ago

Why did I start training from epoch=96 after loading the pre-trained model to start training? Thank you

jorgepradoh commented 1 year ago

Why did I start training from epoch=96 after loading the pre-trained model to start training? Thank you

This happens when you use a checkpoint, epoch=96 is most likely the epoch where the pre-trained model got the best weights.