umautobots / driving-in-the-matrix Goto Github PK

Hello,

I have some problems in reproducing the result of this paper. Could you clarify the learning rate schedule for this paper? Thanks a lot!

In the paper, it mentioned, "Training began with a learning rate of 1e-3 and decreased by a factor of 10 after every 10k iterations until a minimum learning rate of 1e-8 was achieved."

My questions are:

1. The learning rate schedule in mxnet is defined in epochs not iterations. At least the file incubator-mxnet/example/rcnn/train_end2end.py doesn't accept iterations auguments.

2. Image flipping for data augmentation is on by default in mxnet, which means 1 epoch of 10k images contains 20k iterations. Is this definition of iteration match the one used in paper? Or it should be off.

3. Starting from 1e-3, every 10k iteration, learning rate decreased by 10 until 1e-8. That means most later samples in 200k data set will be used when the learning rate is very low (1e-8). Would that be a problem that these samples are more of less ignored in trainning? Due to very low learning rate.

If possible, could you also provide the file of training parameters setting or other code change of mxnet rcnn? That would make the last bit clear when reproducing results of this paper.

PS, I didn't use the docker environment, due to some gcc issues. But I think it's not the problem, since all changes in that docker are already merged into mainline.

Thank you very much!

-Liang

Hello, do you also render the depth images? I saw a stereo camera somewhere in the GTA code, but I cannot find any images of it. Sorry if I missed them...

Is it possible to modify the class list of the provided data? I would like to add, for example, classes for stopSign and pedestrianCrossingSign. I saw https://github.com/umautobots/GTAVisionExport/tree/master/native which would allow me to capture/generate new synthetic data, but I am specifically wondering if the existing data can be modified programmatically to have annotations for new classes.

Thanks for this great work!

Just curious as to the progress...

(https://fcav.engin.umich.edu/projects/driving-in-the-matrix) It seems that the data set in this link is not normal. The download failure prompt will always appear next week.

Dataset link is not valid. Can someone please confirm?

Hello,

i am trying to follow the build/train/evaluate instructions but get stuck at the building docker image stage:
...
++ -std=c++11 -c -DMSHADOW_FORCE_STREAM -Wall -O3 -I/root/mxnet/mshadow/ -I/root/mxnet/dmlc-core/include -fPIC -I/root/mxnet/nnvm/include -Iinclude -funroll-loops -Wno-unused-variable -Wno-unused-parameter -Wno-unknown-pragmas -Wno-unused-local-typedefs -msse3 -I/usr/l
ocal/cuda/include -DMSHADOW_USE_CBLAS=1 -DMSHADOW_USE_MKL=0 -DMSHADOW_RABIT_PS=0 -DMSHADOW_DIST_PS=0 -DMSHADOW_USE_PASCAL=0 -DMXNET_USE_OPENCV=1 -I/usr/include/opencv -fopenmp -DMSHADOW_USE_CUDNN=1 -I/root/mxnet/cub -DMXNET_USE_NVRTC=0 -MMD -c src/io/io.cc -o build/src
/io/io.o
In file included from src/io/./iter_batchloader.h:16:0,
from src/io/io.cc:7:
src/io/././inst_vector.h:157:46: error: expected class-name before '{' token
class TBlobContainer : public mshadow::TBlob {
^
src/io/././inst_vector.h:166:30: error: 'TShape' in namespace 'mshadow' does not name a type
void resize(const mshadow::TShape &shape, int type_flag) {
^
src/io/././inst_vector.h: In constructor 'mxnet::io::TBlobContainer::TBlobContainer()':
src/io/././inst_vector.h:160:21: error: expected class-name before '(' token
: mshadow::TBlob(), tensor_container_(nullptr) {}

Please help.

Thank you in advance!

Hi All,

I would like to get your training pipeline for benchmarking up and running. However, I'm facing the following issue while building the MXNet RCNN container:

`You are in 'detached HEAD' state. You can look around, make experimental
changes and commit them, and you can discard any commits you make in this
state without impacting any branches by performing another checkout.

If you want to create a new branch to retain commits you create, you may
do so (now or later) by using -b with the checkout command again. Example:

git checkout -b

HEAD is now at 5568641... fixes loading of voc evaluations for empty classes (#4840)
Makefile:27: mshadow/make/mshadow.mk: No such file or directory
Makefile:28: /root/mxnet/dmlc-core/make/dmlc.mk: No such file or directory
Makefile:126: /root/mxnet/ps-lite/make/ps.mk: No such file or directory
make: *** No rule to make target '/root/mxnet/ps-lite/make/ps.mk'. Stop.
The command '/bin/sh -c cd /root && git clone --recursive https://github.com/dmlc/mxnet && cd mxnet && git checkout 5568641d99c7d7dac2aaab53d35f0a70c15b3a7f && cp make/config.mk config.mk && sed -i 's/USE_BLAS = atlas/USE_BLAS = openblas/g' config.mk && sed -i 's/USE_CUDA = 0/USE_CUDA = 1/g' config.mk && sed -i 's/USE_CUDA_PATH = NONE/USE_CUDA_PATH = /usr/local/cuda/g' config.mk && sed -i 's/USE_CUDNN = 0/USE_CUDNN = 1/g' config.mk && sed -i 's/EXTRA_OPERATORS =/EXTRA_OPERATORS = example/rcnn/operator/g' config.mk && sed -i 's/-gencode arch=compute_50,code=compute_50/-gencode arch=compute_50,code=compute_50 -gencode arch=compute_60,code=sm_60 -gencode arch=compute_61,code=sm_61 -gencode arch=compute_62,code=sm_62/g' config.mk && make -j"$(nproc)"' returned a non-zero code: 2
`
Do you have any ideas / suggestions how to fix this? I'm getting the same error on two different Linux machines....

Despite that I would like to run your pipeline outside of docker. Can I find a guidance for this somewhere? Looking at the docker guidance I cannot figure out how to run the training... It is unclear to me where train_end2end.py script is coming from. I don't see this in the Faster R-CNN example in MXNet under incubator-mxnet/example/rcnn/. The existing train.py script does not run somehow on your data...

Thanks in advance for your help! Any suggestions are highly appreciated!

Best,
Alexey