Filtered roidb entries to 0

[zuishusheng]

when I train the dataset，gives the error tips

And I fix the problem, because roidb is filtered.

Why the roidb is filtered? Thanks

[pengzhou1108]

when I train the dataset，gives the error tips

And I fix the problem, because roidb is filtered.

Why the roidb is filtered? Thanks

Hi,

As the comments indicate, this step makes sure that the training images are valid. (e.g. it has bounding box annotation so that RoIs are exist)

[zuishusheng]

when I train the dataset，gives the error tips

And I fix the problem, because roidb is filtered.


Why the roidb is filtered? Thanks

Hi,

As the comments indicate, this step makes sure that the training images are valid. (e.g. it has bounding box annotation so that RoIs are exist)

sorry,I don't understand exactly. Is there a problem with the dataset I generated?
There only pictures,after run run_demo.sh.

And the roid info detail .

The 'max_overlaps' array is null.
Can you help me figure out the problem more detailed? And how to solve it?
Thanks!

[pengzhou1108]

when I train the dataset，gives the error tips

And I fix the problem, because roidb is filtered.


Why the roidb is filtered? Thanks

Hi,
As the comments indicate, this step makes sure that the training images are valid. (e.g. it has bounding box annotation so that RoIs are exist)

sorry,I don't understand exactly. Is there a problem with the dataset I generated?
There only pictures,after run run_demo.sh.

And the roid info detail .

The 'max_overlaps' array is null.
Can you help me figure out the problem more detailed? And how to solve it?
Thanks!

Hi,
It seems you didn't read the bounding box annotation in train_filter_single.txt. Have you run split_train_test.py after run_demo.sh to generate txt file? I have updated the code to make it more clear, hope it helps. (you have to change path variable in split_train_test.py)

[hengxiangYao]

when I train the dataset，gives the error tips

And I fix the problem, because roidb is filtered.


Why the roidb is filtered? Thanks

Hi,
As the comments indicate, this step makes sure that the training images are valid. (e.g. it has bounding box annotation so that RoIs are exist)

sorry,I don't understand exactly. Is there a problem with the dataset I generated?
There only pictures,after run run_demo.sh.

And the roid info detail .

The 'max_overlaps' array is null.
Can you help me figure out the problem more detailed? And how to solve it?
Thanks!

Hi,
It seems you didn't read the bounding box annotation in train_filter_single.txt. Have you run split_train_test.py after run_demo.sh to generate txt file? I have updated the code to make it more clear, hope it helps. (you have to change path variable in split_train_test.py)

Hi,
I encountered the exact same issue.
But I did run the split_train_test.py to generate txt file. The file it generated named train_filter.txt, I assume that should be train_filter_single.txt and modify its name. But the same problem occured,

If relevent,

Can you help me?
Thank you very much!

[pengzhou1108]

when I train the dataset，gives the error tips

And I fix the problem, because roidb is filtered.


Why the roidb is filtered? Thanks

Hi,
As the comments indicate, this step makes sure that the training images are valid. (e.g. it has bounding box annotation so that RoIs are exist)

sorry,I don't understand exactly. Is there a problem with the dataset I generated?
There only pictures,after run run_demo.sh.

And the roid info detail .

The 'max_overlaps' array is null.
Can you help me figure out the problem more detailed? And how to solve it?
Thanks!

Hi,
It seems you didn't read the bounding box annotation in train_filter_single.txt. Have you run split_train_test.py after run_demo.sh to generate txt file? I have updated the code to make it more clear, hope it helps. (you have to change path variable in split_train_test.py)

Hi,
I encountered the exact same issue.
But I did run the split_train_test.py to generate txt file. The file it generated named train_filter.txt, I assume that should be train_filter_single.txt and modify its name. But the same problem occured,

If relevent,

Can you help me?
Thank you very much!

Hi,

Do you have bounding box annotations in the txt files? You can check if the bbox looks correct or not.

[hengxiangYao]

when I train the dataset，gives the error tips

And I fix the problem, because roidb is filtered.


Why the roidb is filtered? Thanks

Hi,
As the comments indicate, this step makes sure that the training images are valid. (e.g. it has bounding box annotation so that RoIs are exist)

sorry,I don't understand exactly. Is there a problem with the dataset I generated?
There only pictures,after run run_demo.sh.

And the roid info detail .

The 'max_overlaps' array is null.
Can you help me figure out the problem more detailed? And how to solve it?
Thanks!

Hi,
It seems you didn't read the bounding box annotation in train_filter_single.txt. Have you run split_train_test.py after run_demo.sh to generate txt file? I have updated the code to make it more clear, hope it helps. (you have to change path variable in split_train_test.py)

Hi,
I encountered the exact same issue.
But I did run the split_train_test.py to generate txt file. The file it generated named train_filter.txt, I assume that should be train_filter_single.txt and modify its name. But the same problem occured,

If relevent,

Can you help me?
Thank you very much!

Hi,

Do you have bounding box annotations in the txt files? You can check if the bbox looks correct or not.

Hi, do you mean train_filter_single.txt?

It looks like that. Does that contain any bbox annotations?

[hengxiangYao]

Hi,
I use the synthetic dataset you provide and it worked.
Now I got another error while training:

E tensorflow/stream_executor/cuda/cuda_fft.cc:169] failed to create cuFFT batched plan:2
E tensorflow/stream_executor/cuda/cuda_fft.cc:111] failed to run cuFFT routine cufftSetStream: 1
I tensorflow/stream_executor/stream.cc:1338] stream 0x561fac6d59b0 did not wait for stream: 0x561fac6d4db0
I tensorflow/stream_executor/stream.cc:3788] stream 0x561fac6d59b0 did not memcpy host-to-device; source: 0x1041ce00000
F tensorflow/core/common_runtime/gpu/gpu_util.cc:343] CPU->GPU Memcpy failed
./train_faster_rcnn.sh: line 97: 32386 Aborted (core dumped) python3 ./tools/trainval_net.py --weight ./data/resnet_v1_101.ckpt --imdb ${TRAIN_IMDB} --imdbval ${TEST_IMDB} --iters ${ITERS} --cfg cfgs/${NET}.yml --tag ${EXTRA_ARGS_SLUG} --net ${NET} --set ANCHOR_SCALES ${ANCHORS} ANCHOR_RATIOS ${RATIOS} TRAIN.STEPSIZE ${STEPSIZE} ${EXTRA_ARGS}

I read another issue that has similar problem, you says that this is due to incorrect way to change the RPN batch size and we need to change it in cofg, and I did change it to 1 or 2 in config/res101_fusion.yml, but the error still occur.
Do I need to modify other setting as well?
Thanks in advance!