Open anguszxd opened 5 years ago
No problem, I think 70 percent is probably acceptable with what you are describing. What training accuracy did you reach? I have a presentation with some more details about our project, I can send that to you if you'd like.
On Tue, Jan 15, 2019, 5:18 PM anguszxd <notifications@github.com wrote:
Hi @ms3001 https://github.com/ms3001 , thanks for your code. I am trying this project, and it's very simple to apply. But when I try to train C3D model, I find the result seems to be odd. The training loss can't reduce during training process, and the accuracy is very low. I'm using config file config_C3D_default.json . ConvColumn model trains well, I train convcolumn net 30 epoches, and the test accuray is around 70%, I think this result is right. Have you ever tried to train C3D model with 20BN dataset ? Thanks.
— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/ms3001/DeepHandGestureRecognition/issues/1, or mute the thread https://github.com/notifications/unsubscribe-auth/ALg_mYZrjZpwDEThM0wFHdShC4WRAfJvks5vDn4CgaJpZM4aCH99 .
Hi @ms3001 , thanks for you reply! When I train C3D model with NV Gestures dataset, the training loss never reduces during the whole process, so does the accuracy. Would you please send me the presentation ? It will be very helpful. Thanks very! My email address: zxd0520@163.com
Hi @ms3001 , thanks for your code. I am trying this project, and it's very simple to apply. But when I try to train C3D model, I find the result seems to be odd. The training loss can't reduce during training process, and the accuracy is very low. I'm using config file
config_C3D_default.json
. ConvColumn model trains well, I train convcolumn net 30 epoches, and the test accuray is around 70%, I think this result is right. Have you ever tried to train C3D model with 20BN dataset ? Thanks.