Unable to Reproduce DCASE2019 Task4 challenge results

Here’s the error log -

2020-01-15 07:20:51.994231: I tensorflow/core/platform/cpu_feature_guard.cc:140] Your CPU supports instructions that this TensorFlow binary was not compiled to use: SSE4.1 SSE4.2 AVX AVX2 FMA
2020-01-15 07:20:51.998584: E tensorflow/stream_executor/cuda/cuda_driver.cc:406] failed call to cuInit: CUDA_ERROR_NO_DEVICE
2020-01-15 07:20:51.998641: I tensorflow/stream_executor/cuda/cuda_diagnostics.cc:158] retrieving CUDA diagnostic information for host: instance-2
2020-01-15 07:20:51.998654: I tensorflow/stream_executor/cuda/cuda_diagnostics.cc:165] hostname: instance-2
2020-01-15 07:20:51.998707: I tensorflow/stream_executor/cuda/cuda_diagnostics.cc:189] libcuda reported version is: 440.33.1
2020-01-15 07:20:51.998765: I tensorflow/stream_executor/cuda/cuda_diagnostics.cc:193] kernel reported version is: 440.33.1
2020-01-15 07:20:51.998779: I tensorflow/stream_executor/cuda/cuda_diagnostics.cc:300] kernel version seems to match DSO: 440.33.1
 INFO - Disentangled feature
 INFO - task name: Lin_ICT_task4
 INFO - PS-model name: sed_with_cATP-DF
 INFO - mode: test
 INFO - semi_supervised: False
 INFO - ensemble results (just a single model)
Traceback (most recent call last):
  File "main.py", line 307, in <module>
    test_models(task_name, sed_model_name, model_weights_list)
  File "main.py", line 202, in test_models
    test(task_name, sed_model_name, model_list[0])
  File "main.py", line 145, in test
    train = trainer.trainer(task_name,model_name,True)
  File "/home/udaylunawat/Sound_event_detection/src/trainer.py", line 76, in __init__
    self.init_data()
  File "/home/udaylunawat/Sound_event_detection/src/trainer.py", line 128, in init_data
    self.data_loader = data.data_loader(conf_dir)
  File "/home/udaylunawat/Sound_event_detection/src/data_loader.py", line 49, in __init__
    self.init_data_conf()
  File "/home/udaylunawat/Sound_event_detection/src/data_loader.py", line 88, in init_data_conf
    assert os.path.exists(f)
AssertionError

I’ve checked the tensorflow-gpu installation & it works.

>>> tf.test.is_gpu_available()
2020-01-15 07:20:44.186951: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1435] Adding visible gpu devices: 0
2020-01-15 07:20:44.187094: I tensorflow/core/common_runtime/gpu/gpu_device.cc:923] Device interconnect StreamExecutor with strength 1 edge matrix:
2020-01-15 07:20:44.187108: I tensorflow/core/common_runtime/gpu/gpu_device.cc:929]      0 
2020-01-15 07:20:44.187116: I tensorflow/core/common_runtime/gpu/gpu_device.cc:942] 0:   N 
2020-01-15 07:20:44.187209: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1053] Created TensorFlow device (/device:GPU:0 with 244 MB memory) -> physical GPU (device: 0, name: Tesla K80, pci bus id: 0000:00:04.0, compute capability: 3.7)
True

Would you also tell me the directory structure for the data/wav and data/feature folder. Thanks!!

[ckdals7419]

Hi, I am also trying to reproduce this model that won first place in DCASE 2019.

Like you answered above,

I uploaded all the downloaded files to data/wav, and finished data preparation by running gen_feature-2019.sh and gen_label-2019.sh

Then I ran sh scripts/reproduce_Lin_ICT_task4_*.sh and an error occurred in the load_weights function.

This is error log.

(venv) changmin@kresnik:~/Sound_event_detection$ sh scripts/reproduce_Lin_ICT_task4_4.sh Using TensorFlow backend. 2020-01-16 23:23:37.829291: I tensorflow/core/platform/cpu_feature_guard.cc:140] Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX2 AVX512F FMA INFO - Disentangled feature INFO - task name: Lin_ICT_task4 INFO - PS-model name: sed_with_cATP-DF INFO - mode: test INFO - semi_supervised: False INFO - decode for model : challenge_results/model_weights/437296/best_model_w.h5 Traceback (most recent call last): File “main.py”, line 309, in <module> test_models(task_name, sed_model_name, model_weights_list) File “main.py”, line 210, in test_models at_preds, sed_preds = test(task_name, sed_model_name, model_path) File “main.py”, line 149, in test at_preds_out = train.save_at_result(at_preds) #audio tagging result File “/mnt/sdb1/changmin/Sound_event_detection/src/trainer.py”, line 372, in save_at_result preds_out[‘vali’] = self.save_at(‘vali’, at_preds, is_add = False) File “/mnt/sdb1/changmin/Sound_event_detection/src/trainer.py”, line 413, in save_at preds, labels = self.test(mode, ‘at’, at_preds) File “/mnt/sdb1/changmin/Sound_event_detection/src/trainer.py”, line 337, in test mode = mode) File “/mnt/sdb1/changmin/Sound_event_detection/src/model.py”, line 596, in get_model model.load_weights(pre_model, by_name = True) File “/home/changmin/venv/lib/python3.6/site-packages/keras/engine/network.py”, line 1177, in load_weights reshape=reshape) File “/home/changmin/venv/lib/python3.6/site-packages/keras/engine/saving.py”, line 1018, in load_weights_from_hdf5_group_by_name str(weight_values[i].shape) + ‘.’) ValueError: Layer #87 (named “sed_with_cATP-DF_atten_0”), weight <tf.Variable ‘sed_with_cATP-DF_atten_0/sed_with_cATP-DF_atten_0_kernel:0’ shape=(1, 1, 138) dtype=float32_ref> has shape (1, 1, 138), but the saved weight has shape (1, 1, 137).

Previously saved challenge_results/model_weights/437296/best_model_w.h5 I just loaded the model, and I’m embarrassed to see this error.

I really want to study using this model. I would be grateful if you could help. thank you : )

[Kikyo-16]

Hi, the folder data/wav contains all the audio files published by DCASE2019 Task4 (no any subfolder). Before running sh scripts/reproduce_Lin_ICT_task4_*.sh to reproduce the results, you should run gen_feature-2019.sh and gen_label-2019.sh to generated features and labels.