search

qinche106 / cb-convlstm-eyetracking

Event-based Eye Tracking using ConvLSTM networks
MIT License
19 stars 4 forks source link

Failing to replicate the right result #6

Closed HJL8055 closed 10 months ago

HJL8055 commented 11 months ago

Hello,

I am trying to run the model. Since I don't know how to handle with the path(/DATA/pupil_st/data_ts_500) in the file"process_event.py", I skip the second step and run convlstm-et-pytorch-event.py with the data which is downloaded from Synthetic Event-based Eye Tracking (SEET) dataset link (event frame).

But after training 60 epochs, I get a bad result.

Size 60, Epoch 59, Loss: 0.007099685608409345, err_rate_1:0.9864583611488342, err_rate_3:0.8781250715255737, err_rate_5:0.6864583492279053 err: 0.33020836114883423 num_values: 317 tot_values: 960 event_plot_59 I appreciate any assistance or insights into resolving this issue. Thank you in advance!

ZuowenWang0000 commented 10 months ago

Hello, Could you please re-download the dataset and also pull the code again and try again? We uploaded a validation file with the wrong stride, which caused a mismatch with the ground truth. It should be fixed now. Please be aware in these plots, there are two recordings, each with 480 data points.

Sorry for causing the inconvenience!

HJL8055 commented 10 months ago

Hello, Could you please re-download the dataset and also pull the code again and try again? We uploaded a validation file with the wrong stride, which caused a mismatch with the ground truth. It should be fixed now. Please be aware in these plots, there are two recordings, each with 480 data points.

Sorry for causing the inconvenience!

Wow, it worked for me! Thank you very much! event_plot_84