Where to find ImageSets/train.txt

[klockeph]

Hi!

After a long journey I finally got this code to work (can document some steps if someone is interested) but when starting the training it says Path does not exist: [...]/MV3D_TF/tools/../lib/datasets/../../data/KITTI/ImageSets/train.txt.

Where should this file come from?

Thanks!

[Durant35]

OK, I will try to bringup this repo soon, with runnable README, not just fork...
Maybe the leading 2 are better?

[klockeph]

I think actually this one is the best fork at the moment.
The other interesting one seems to be the one you mentioned from bostondiditeam. I couldn't try it though because I have no GPU currently

lmb-freiburg used caffee and leeyevi is behind your repo by 1 commit.

[Durant35]

Hi!

After a long journey I finally got this code to work (can document some steps if someone is interested) but when starting the training it says Path does not exist: [...]/MV3D_TF/tools/../lib/datasets/../../data/KITTI/ImageSets/train.txt.

Where should this file come from?

Thanks!

I also met this issue, and following https://github.com/sshaoshuai/PointRCNN/blob/master/data/KITTI/ImageSets/train.txt, directly copied it to $MV3D/data/KITTI/ImageSets/train.txt work for me. I'm making a runnable version based on the previous author.

(python2.7-tf1.0.0) gary@.....:~/Workspace/git_ws/MV3D(runnable_README)$ ./experiments/scripts/mv3d.sh gpu 0 data/pretrain_model/VGG_imagenet.npy kitti_train
...
iter: 360 / 50001, total loss: 0.2785, rpn_loss_cls: 0.0542, rpn_loss_box: 0.0262, loss_cls: 0.0975, loss_box: 0.1007, lr: 0.000010
speed: 1.330s / iter
iter: 370 / 50001, total loss: 0.8002, rpn_loss_cls: 0.1742, rpn_loss_box: 0.2732, loss_cls: 0.3004, loss_box: 0.0524, lr: 0.000010
speed: 1.330s / iter
iter: 380 / 50001, total loss: 0.8278, rpn_loss_cls: 0.0968, rpn_loss_box: 0.3903, loss_cls: 0.1282, loss_box: 0.2125, lr: 0.000010
speed: 1.329s / iter
iter: 390 / 50001, total loss: 0.8136, rpn_loss_cls: 0.1375, rpn_loss_box: 0.3292, loss_cls: 0.2142, loss_box: 0.1327, lr: 0.000010
speed: 1.328s / iter
iter: 400 / 50001, total loss: 1.0774, rpn_loss_cls: 0.1111, rpn_loss_box: 0.3365, loss_cls: 0.2336, loss_box: 0.3962, lr: 0.000010
speed: 1.327s / iter
iter: 410 / 50001, total loss: 1.0331, rpn_loss_cls: 0.1357, rpn_loss_box: 0.4888, loss_cls: 0.1287, loss_box: 0.2799, lr: 0.000010
speed: 1.326s / iter
iter: 420 / 50001, total loss: 0.4182, rpn_loss_cls: 0.0889, rpn_loss_box: 0.2045, loss_cls: 0.0401, loss_box: 0.0847, lr: 0.000010
speed: 1.326s / iter
...

PS: 1.326s / iter means 18h totally for training...