ankitshah009 / task-4-large-scale-weakly-supervised-sound-event-detection-for-smart-cars Goto Github PK

Hi,

I'm getting zero scores when running evaluation/TaskAEvaluate.py due to newline not being correctly removed from groundtruth and predictions labels.

I suggest changing line 19 in evaluation/Models.py to label = line.split("\t")[3].strip() to better handle whitespace characters. May also need to add somewhere for predicted_labels.

Thanks.

Running with defaults on Scientific Linux 6.9, I am getting this encoding error when dowloading the files

[I]   Evaluator

[I]     Save path            : system/task4/evaluator

Files                    :   0%|                      | 0/51660 [00:00<?, ?it/s]

'latin-1' codec can't encode characters in position 33-37: ordinal not in range(256)

Edit, solved...

Hey,

In line 73 of file task4.py there is the code:

sed_eval.sound_event.SegmentBasedMetrics

There is no package named sed_eval in the task4 framework nor in the sound event framework. How can one replace this sed_eval package? Or, where one can find this package?

The same stands for line 79.

Best,

sudo apt-get install ffmpeg

I downloaded the data but I don't know where to put them and so that I can skip the download_data process. Does anyone know that?

Why is there so many utterance without any label (totally empty)???

For example, the utterance starts from zero and ends also at zero, what does it mean? The problem is my trained model can always detect something, but the label is empty, leading to more insertion errors.

hVvtTC9AmNs_30.000_40.000.wav 0 0 Train

Hello,I want to know, I have files in a submitted form,but I don't know how to use the code in public. It doesn't work