deep learning- final project clarifications

ForumCategory: Deep Learningdeep learning- final project clarifications
Chen Hazut asked 1 month ago
  1. From what I understand of your previous answer in the forum we need to have 5 classes at least. This means 5 different classes that come from the world of traffic management?
  2. I intend to use the mobileNet pre-trained model. Do I need to train it on two other datasets or is one enough?
  3. The mobileNet already contains the classes I want to use (like traffic lights, cars signs etc.) Can the different dataset contain the same classes but with different pictures?
  4. How many images needs to be in the other dataset?
  5. What is needed to be in the code we write? I understand that the dataset needs to be prepared and the model have to be changed by removing the last layers and replacing them with new ones. But is that all we have to do?
  6. What is need to be written in the report you requested?
  7. Are we allowed to choose a detection task instead of classification\segmentation?  
Comments ( 1 )
  • The OpTeamizer Admin says:

    1. Yes.
    2. Need to train on two
    3. The different dataset should contain the same classes with different pictures.
    4. Enough to make it accurate. If it is a separate dataset then for sure it has enough images in each of its classes.
    5. You’d probably need to adapt the inputs/output of the different datasets as well.
    6. Description of each of the steps, which data sets you chose, what adaptation you needed to make, the precision that you’re getting for the classes, possible alternatives that you considered while developing, etc.
    7. No

Leave A Comment

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.

1 Answers
Chen Hazut answered 1 month ago

Hello,
Can you please clarify the following issues for me:
We used the mobileNet model which has already been trained on the imageNet dataset. and trained it again on cifar100. 
1.Does the first training on the imageNet dataset counts as one of the two datasets we need to train on? or in addition to the imageNet we need to train on two other datasets?
2.In case we need to train on two other dataset can the first contain images from imageNet and the second  contain images from cifar100 for eample? or do we need to find a third dataset?
3. ln addition, in case we need to use two datasets, do we need to combine them into one dataset and train the model once or can we train the model on the first dataset and then on the other one?
thanks

Comments ( 1 )
  • The OpTeamizer Admin says:

    1. “Does the first training on the imageNet dataset counts as one…”
    The first counts as one
    2.
    3. You don’t have to merge them.

Leave A Comment

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.

Comments ( 1 )
  • The OpTeamizer Admin says:

    1. Yes.
    2. Need to train on two
    3. The different dataset should contain the same classes with different pictures.
    4. Enough to make it accurate. If it is a separate dataset then for sure it has enough images in each of its classes.
    5. You’d probably need to adapt the inputs/output of the different datasets as well.
    6. Description of each of the steps, which data sets you chose, what adaptation you needed to make, the precision that you’re getting for the classes, possible alternatives that you considered while developing, etc.
    7. No

The comments are now closed.