Soda Bottle Identification Challenge


#42

Even though I got a very high accuracy on my training and validation set, I got a score of Zero??
I don’t think that it’s even possible that my model classified everything wrong.
Even if some random model classifies those images, still there would 12.5% accuracy.


#43

Is there any way I could know if the submission reached you? At least a mail stating that your entry is received would be cool! I submitted and I am still wondering if the previous submission was successful and I would suggest the evaluation process to be automated as well because we should know that we need to improve the model immediately :smiley:


#44

I think you need to resize the image. When the computation becomes heavy and crashes then it shows this error. After Resize I did not encountered this error!


#45

@Harveen_chadha
@Multitude

  • It’s a challenge. ™

Bad data, little power, limiter time, hunger, holidays, I want to drink I want to sleep I want rest

Willkommen!!! =D


#46

I have the same issues uploading bigger models. There is no indicator about progress and after a while it shows [object object] and not Sucessfully uploaded.

Is there a size limit for the uploaded models?

Also I only got a confirmation mail for the first model submitted. The Second one is showing as it says 2 submission. But no information about the score and no email at all.


#47

After training a model, we need to download it from your server first and then upload it to your server.
It’s so wired. It wasted so much network and time.
Why not let us submit a url link to our model directly?


#48

it’s been more than 10 hours but still i’m unable to see myscore on leaderboard, anybody facing the same issue.


#49

Do you train the model again after we submit it in zip file and then use it for prediction or you directly use the model trained by us in zip file i.e that .h5 file


#51

i trained the model, after that i went to inference and downloaded the model in zip file which consisted of .h5 file [Problem solved]


#52

@deepsea,

We dont have access to trained models in case users are using Desktop version. So currently we require all users to follow the same submission instructions.

But I understand your point and we may enable some optimization in submission process in future competitions.


#53

@rajendra Hi sir. Could you please say me that if you use our trained model(.h5 file,present in zip file). I am asking this because I got a score of zero 4 time even though I have very high accuracy


#54

@ManikSoni yes, we use the trained model (.h5 file in zip file) for evaluation. If you are getting zero score, that means some error occurred during evaluation with your model.

You can try a very basic model first and run for at least one epoch and try uploading. Hopefully, that should generate some score. and you can optimize further from there.

We will check our server logs in the meantime to understand why your model is failing.


#55

@ManikSoni, no we do not do further training. we use the trained model for evaluation.


#56

Please do check the logs sir


#57

@ManikSoni your model is causing out of memory error on the system. You should resize the images to smaller resolution.


#58

How to use the reshape block?
I added a tuple for e.g. (100,100,3)
But it showed the following error
ValueError : total size of new array must be unchanged


#59

I have uploaded my model for many times, it still gives a hint: [object Object] or error


#60

I resized them, accuracy is also good but still 0 points. Help


#61

Can you upload your model to submit it?
I cannot upload my model successfully, and it gives a tip: [object Object]


#63

I have made 2 submissions and none show up on the LB. I checked and my zip file has .h5 model. Can someone let me know what i might be doing wrong?