search

LossNAN / I3D-Tensorflow

Train I3D model on ucf101 or hmdb51 by tensorflow
Apache License 2.0
112 stars 28 forks source link

Can you offer me a copy of the trained model on all the 3 splits of UCF101 and HMDB ,Thank you! #2

Open lzylzy2223 opened 5 years ago

lzylzy2223 commented 5 years ago

Have you trained the model on all 3 splits of the UCF101 and HMDB ? If you trained,can you offer me a copy of it,the paper dealine is approaching and I have no time to train all the 3 splits, Thank you!

LossNAN commented 5 years ago

You can use the kinetics pre-train model in checkpionts, they are better than ucf or hmdb, and finetuning ucf101 or hmdb on kinetics-pretrain just use less than half day(one gpu)

lzylzy2223 notifications@github.com 于2018年10月10日周三 下午6:48写道：

Have you trained the model on all 3 splits of the UCF101 and HMDB ? If you trained,can you offer me a copy of it,the paper dealine is approaching and I have no time to train all the 3 splits, Thank you!

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/LossNAN/I3D-Tensorflow/issues/2, or mute the thread https://github.com/notifications/unsubscribe-auth/ARnvXxdD1MwVCYmChH0o54d91YrmJ10aks5ujdBvgaJpZM4XVEjo .

lzylzy2223 commented 5 years ago

You can use the kinetics pre-train model in checkpionts, they are better than ucf or hmdb, and finetuning ucf101 or hmdb on kinetics-pretrain just use less than half day(one gpu) lzylzy2223 notifications@github.com 于2018年10月10日周三 下午6:48写道： … Have you trained the model on all 3 splits of the UCF101 and HMDB ? If you trained,can you offer me a copy of it,the paper dealine is approaching and I have no time to train all the 3 splits, Thank you! — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#2>, or mute the thread https://github.com/notifications/unsubscribe-auth/ARnvXxdD1MwVCYmChH0o54d91YrmJ10aks5ujdBvgaJpZM4XVEjo .

Thank you for answering,I trained this model for 20000 step at first but I found the result is not very good so I trained them 40 epoch every dataset. But just now, I found the 40 epoch result is just like the 20000 step result.Maybe It is not the reason. So now I have time to train this model,HAHA~!

LossNAN commented 5 years ago

You are welcome!

lzylzy2223 notifications@github.com 于2018年10月10日周三 下午7:16写道：

You can use the kinetics pre-train model in checkpionts, they are better than ucf or hmdb, and finetuning ucf101 or hmdb on kinetics-pretrain just use less than half day(one gpu) lzylzy2223 notifications@github.com 于2018年10月10日周三 下午6:48写道： … <#m8236134685488318214> Have you trained the model on all 3 splits of the UCF101 and HMDB ? If you trained,can you offer me a copy of it,the paper dealine is approaching and I have no time to train all the 3 splits, Thank you! — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#2 https://github.com/LossNAN/I3D-Tensorflow/issues/2>, or mute the thread https://github.com/notifications/unsubscribe-auth/ARnvXxdD1MwVCYmChH0o54d91YrmJ10aks5ujdBvgaJpZM4XVEjo .

Thank you for answering,I trained this model for 20000 step at first but I found the result is not very good so I trained them 40 epoch every dataset. But just now, I found the 40 epoch result is just like the 20000 step result.Maybe It is not the reason. So now I have time to train this model,HAHA~!

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/LossNAN/I3D-Tensorflow/issues/2#issuecomment-428534157, or mute the thread https://github.com/notifications/unsubscribe-auth/ARnvXx_ooS9Bkhm76d2QwNfYu1fZqr7oks5ujdcigaJpZM4XVEjo .