tensorflow
build_image_dataset.py crashes for delf training
Running instructions from https://github.com/tensorflow/models/blob/master/research/delf/delf/python/training/README.md#prepare-the-data-for-training without any GPUs on a Google Cloud VM encounters an error.
Below is the command with the error
python3 build_image_dataset.py --train_csv_path=$LANDMARK_DATA/train/train.csv --train_clean_csv_path=$LANDMARK_DATA/train/train_clean.csv --train_directory=$LANDMARK_DATA/train//// --output_directory=$LANDMARK_DATA/tfrecord/ --num_shards=128 --generate_train_validation_splits --validation_split_size=0.2 --test_csv_path=$LANDMARK_DATA/train/test.csv --test_directory=$LANDMARK_DATA/test////
2022-01-27 22:07:03.277440: W tensorflow/stream_executor/platform/default/dso_loader.cc:64] Could not load dynamic library 'libcudart.so.11.0'; dlerror: libcudart.so.11.0: cannot open shared object file: No such file or directory
2022-01-27 22:07:03.277495: I tensorflow/stream_executor/cuda/cudart_stub.cc:29] Ignore above cudart dlerror if you do not have a GPU set up on your machine.
2022-01-27 22:07:05.252430: W tensorflow/stream_executor/platform/default/dso_loader.cc:64] Could not load dynamic library 'libcuda.so.1'; dlerror: libcuda.so.1: cannot open shared object file: No such file or directory
2022-01-27 22:07:05.252502: W tensorflow/stream_executor/cuda/cuda_driver.cc:269] failed call to cuInit: UNKNOWN ERROR (303)
2022-01-27 22:07:05.252525: I tensorflow/stream_executor/cuda/cuda_diagnostics.cc:156] kernel driver does not appear to be running on this host (gcsfuse-experiment): /proc/driver/nvidia/version does not exist
/home/avidullu/mldata/cvdfoundation/google-landmark/train/train_clean.csv
Traceback (most recent call last):
File "build_image_dataset.py", line 491, in
https://github.com/tensorflow/models/blob/a033df775262a1a48420649a216a16b687bc39f6/research/delf/delf/python/training/build_image_dataset.py#L143 seems to be using a binary mode for read. On removing the 'b' from here and L116 the script makes progress.
Thank you for your post. We noticed you have not filled out the following field in the issue template. Could you update them if they are relevant in your case, or leave them as N/A? Thanks. What is the top-level directory of the model you are using Have I written custom code OS Platform and Distribution TensorFlow installed from TensorFlow version Bazel version CUDA/cuDNN version GPU model and memory Exact command to reproduce
Responding inline
Q. What is the top-level directory of the model you are using A. models/research/delf
Q. Have I written custom code? A. No. Ran the commands as mentioned in research/delf/delf/python/training/README.md
Q. OS Platform and Distribution A. GCP VM with Debian image. 32 CPU and 128GB RAM
Q. TensorFlow installed from? A. From the script mentioned in /research/delf/INSTALL_INSTRUCTIONS.md
Q. TensorFlow version A. 2.7
Q. Bazel version A. 5.0
Q. CUDA/chDNN version A. NA (no GPU on hardware)
Q. GPU model A. NA (no GPU on hardware)
Q Exact command to reproduce A. python3 build_image_dataset.py --train_csv_path=$LANDMARK_DATA/train/train.csv --train_clean_csv_path=$LANDMARK_DATA/train/train_clean.csv --train_directory=$LANDMARK_DATA/train/ --output_directory=$LANDMARK_DATA/tfrecord/ --num_shards=128 --generate_train_validation_splits --validation_split_size=0.2 --test_csv_path=$LANDMARK_DATA/train/test.csv --test_directory=$LANDMARK_DATA/test/
Thanks for reporting this!
@dan-anghel , I think you wrote this part of the code. Do you remember if the 'b' is really necessary when reading the CSV? What @avidullu reported makes sense to me, although I remember you had run the code several times, which seems to contradict it.
@andrefaraujo
Thanks for reporting this!
@dan-anghel , I think you wrote this part of the code. Do you remember if the 'b' is really necessary when reading the CSV?
The CSV shouldn't be binary. Removing b solves the problem for me.
