Ali Darwish, Ph.D.

Great, I will comment on your questions /remarks soon. I have a self-driving car engineer interview on Tuesday and I'm preparing for it :D I just wanted to say a...

@cocoza4 I finally had the chance to look at your code, I hope it's isn't too late :) The code does look correct. I've tested the order for x/y-min/max on...

@cocoza4 which specific part of the code do you want me to look at?

@Peeranat It looks much better.We need to look at the training accuracy and validation accuracy. I've noticed the biases in the Udacity data sets and it is highly recommended to...

Can you try to change the bbox scale factor in ``` def predict(model, image_array, prior_boxes, original_image_shape, num_classes=21, lower_probability_threshold=.1, iou_threshold=.5, background_index=0, box_scale_factors=[.1, .1, .2, .2]): ``` and see if that improves...

That dataset look good! I don't remember if I have created it or its from the repo, here it is ``` import numpy as np from .preprocessing import preprocess_images from...

@cocoza4 in the sample from `ground-truth.pkl` above, you are suppose to have 1 class more than the classes that you are classifying. That class is for background. SO there is...

@cocoza4 sounds good! I wasn't talking about the datasets, I was asking about how gt_pascal.pkl was created. The code for creating it is incomplete. If that's something you don't wish...

@cocoza4 sorry, I was referring to ground-truth.pkl and not gt_pascal.pkl. I ran groundtruth_builder.ipynb but dataset wasn't defined. Weird, if you are certain its the updated one I'll go ahead and...

@cocoza4 Ok, I found the dataset dictionary, I didn't see it the first time, sorry again about that. So I have modified the code to include background, here it is...