rezaghoddoosian / early-drowsiness-detection Goto Github PK

I want to know how data is shown in all .npy file in python, because I want to understand how you train this model with these .npy files

Using the model files you have provided, I have developed a testing code that generates inference over the videos from your database and I have tested it over video feed from the web camera as well.
Most of the predictions are skewed towards the output '10' -> i.e. Drowsy person.

The results do not seem to align with the results that you have shown in your YouTube video.

Could you please share your testing file code so that we can be sure of the correctness of this pipeline?

I use the dataset and blink_video.py to generate alert.txt, semisleepy.txt, sleepy.txt files.
But the outputs of the preprocessing.py which takes the three txt files as its inputs are different from the npy files provided by author.
Is this a potential issue for me to train workable model ?
Thanks.

After reading your paper "A Realistic Dataset and Baseline Temporal Model for Early Drowsiness Detection" published in the CVPR workshop, I have learned a lot. I have encountered some difficulties when trying to reproduce your results. In the literature, your model3 has a VA of 0.70, but when I trained using the npy files you provided, it only reached a maximum of 0.56. What are the parameters such as your learning rate when training this model?

how can i add my own video and test with this per-exisisting model

when i am running blink_video.py, problem shows as follow:

[INFO] starting video stream thread...
Traceback (most recent call last):
File "blink_video.py", line 588, in
blink_detector(output_file,path)
File "blink_video.py", line 341, in blink_detector
top = tk.Tk()
File "/root/miniconda3/envs/myconda/lib/python3.5/tkinter/init.py", line 1876, in init
self.tk = _tkinter.create(screenName, baseName, className, interactive, wantobjects, useTk, sync, use)
_tkinter.TclError: no display name and no $DISPLAY environment variable

Hello,

I got the following error when I run blink_video.py with latest scikit-learn-versions > 0.23:

[INFO] loading facial landmark predictor...
/usr/lib/python3/dist-packages/requests/init.py:89: RequestsDependencyWarning: urllib3 (1.26.6) or chardet (4.0.0) doesn't match a supported version!
warnings.warn("urllib3 ({}) or chardet ({}) doesn't match a supported "
Traceback (most recent call last):
File "blink_video.py", line 581, in
blink_detector(output_file,path)
File "blink_video.py", line 315, in blink_detector
loaded_svm = pickle.load(open('Trained_SVM_C=1000_gamma=0.1_for 7kNegSample.sav', 'rb'))
ModuleNotFoundError: No module named 'sklearn.svm.classes'

And this one with earlier versions (<=0.23) :

[INFO] starting video stream thread...
Attribute Qt::AA_EnableHighDpiScaling must be set before QCoreApplication is created.
/usr/lib/python3/dist-packages/requests/init.py:89: RequestsDependencyWarning: urllib3 (1.26.6) or chardet (4.0.0) doesn't match a supported version!
warnings.warn("urllib3 ({}) or chardet ({}) doesn't match a supported "
Traceback (most recent call last):
File "blink_video.py", line 581, in
blink_detector(output_file,path)
File "blink_video.py", line 444, in blink_detector
IF_Closed_Eyes = loaded_svm.predict(EAR_series.reshape(1,-1))
File "/home/samber9ellas/.local/lib/python3.8/site-packages/sklearn/svm/_base.py", line 606, in predict
if self.break_ties and self.decision_function_shape == 'ovo':
AttributeError: 'SVC' object has no attribute 'break_ties'

Thank you for sharing your work!

About these lines,

Why are these procedure needed??

When convert the model to pb file, what are the output_node_names ?

I am pleased to see your replication of the RTA-RLDD paper. I am a current student and am in need of the RTA-RLDD dataset for experiments related to fatigue detection. The official dataset download link http://vlm1.uta.edu/~athitsos/projects/drowsiness/ is currently unavailable. I would appreciate it if you could provide me with access to the RTA-RLDD dataset. Thank you very much.

Dear Ghoddoosian,

I am a research assistant from Hong Kong Polytechnic University. We are trying using computer version from drowsiness detection for consturction machinary operator.

I have some questions about the codes, would you please provide some suggestions?
In Training.py, there is a comment satates that start indices are the indices in BlinksTest where the set of blink sequences in each video starts. It is used for VA, BSRE and BSA calculation.
And there is if codes for define the start indices for different parts of datasets, and if I want to train the network on my own datasets, how do I define it? Could you explain it further?

Thanks a lot and look forward to your reply.

Best regards,
Daniel