ngessert / isic2019 Goto Github PK

Dear Nils,

I'm trying to recreate your project for my bachelorthesis to have a running machine learning classifier algorithm to compare it to an active learning classifier algorithm. Sadly I cannot get any data from 2019 via this link. So I had to download datasets for the 2020 challenge and delete additional columns and afterwards one-hot encode it. I assume something is wrong with my csv file i.e. are the only columns 'image', 'sex', 'age_approx', 'anatom_site_general_challenge' and 'diagnosis' ?

I would really appreciate to receive an answer.

With kind regards,

Leon

Hello, thank you for your work!
Would you please provide the checkpoint of the pretrained model for inference?
Thank you, Lucia

Hi
Thank you for a great repo, and congratulations on winning the competition. I read your manuscript, however, I can't find the code snippet of preprocessing (cropping the relevant field of view of an image and color constancy). Could you please help me how to do these preprocessing actually.

Regards,
Azam.

Hi Nils,

I am interested to learn about your solution. Could you be able to share the code to create indices_isic2019.pkl and meta_data_official.pkl files?

I have also noticed the following discrepancy in code vs documentation:

• For the final prediction of test images, CSV file name must contain csvFile
• Subset file name must contain subSet
• Line #213 of ensemble.py assumes 5 instead of len(top_inds) as in line #235

Thanks
Sanjib

Sorry to create a new issue since you haven't responsed to the last issue, I was really confuse about the metric called Balanced Multiclass Accuracy, I read the definition but didn't know for sure what is it and how to calculate it, Could you help me answer these question?

Thank you very much!

(tensorflow1.6) newwen@newwen-ECQ3-0:~/Documents/isic/isic2019-master$ python train.py example 2019.test_effb0_ss gpu0
['/home/newwen/Documents/isic/isic2019-master/images/official/']
/home/newwen/anaconda3/envs/tensorflow1.6/lib/python3.5/site-packages/numpy/core/fromnumeric.py:3257: RuntimeWarning: Mean of empty slice.
out=out, **kwargs)
/home/newwen/anaconda3/envs/tensorflow1.6/lib/python3.5/site-packages/numpy/core/methods.py:161: RuntimeWarning: invalid value encountered in double_scalars
ret = ret.dtype.type(ret / rcount)
nan
Len ham 0
Train
(20252,)
(20252,)
(20257,)
(20257,)
(20251,)
Val
(5079,)
(5079,)
(5074,)
(5074,)
(5080,)
Devices to use: 0
Evaluating on validation set during training.
CV set 0
cuda:0
Balance 9
Traceback (most recent call last):
File "train.py", line 204, in
class_weights = 1.0/np.mean(mdlParams['labels_array'][indices_ham,:8],axis=0)
IndexError: too many indices for array

Hi there, thanks for sharing your script. I am having some trouble with eval.py, as it seems mdlParams['input_size_load'] has neither been defined nor has a value been assigned to it. What does this stands for ? Could you help me please?