One option is to convert counts to rate. Rates are always bound between 0 and 1. For example instead of a count of 100 events, the data could be encoded as a rate of .10 (100 events out of 1,000 opportunities).
This kind of problem is called record linkage (or sometimes entity matching or other variants). The task consists in finding among a list of strings representing entities (persons or organizations) those which represent the same actual entity.
There are two main approaches (which can be combined):
String similarity matching methods. See for example this ...
I'm not sure to understand well your problem.
But many you could try using filter functions from Data Frame in Pandas.
In addition to that, it seems more a code issue rather than a data science one: you might have better result in Stack Overflow for this kind of question.
After getting hands on with the data, it feels ridiculous not to fit on all samples in the split/fold.
In 2D data 'sample==row'. You don't fit on a single sample.
In 3D data 'sample==sequence' so you encode on all of the sequences.
This also means less encoders to keep track of for the sake of inverse transform and inference encoding.
So the answer came to me, as answers are wont to do, in the shower: since we're trying to figure out overall direction here, instead of trying to figure out motion frame by frame, just compare the first and last detected positions of each object.
movement = first['left'] - last['left']
direction = 'right' if movement < 0 else 'left'
if abs(movement) < ...
I suggest you play with the height parameter. According to your figure the peaks are easy to detect. You could use the most frequent value as offset for the height parameter, but I think you should play with those values. After you find the desired peaks, you can simply do a left and right search so that you calculate the average and replace it instead of ...
I'm struggling with a similar problem myself.
Could you please provide details about:
What version of efficientDet you are using
What preprocessing you are currently using on the images
One solution is to divide the images into smaller images. At this years Nvidia GTC conference, ConservationAI did a talk where they mentioned that they split up 8k images ...
One hot encoder is your best choice. Still, you have to deal with enlarged dimensional size, though, as long as you don't drop categories you are not giving preferences.
Dropping is needed to avoid collinearity which is a fancy way to say: "I have four friends, Anne, Bart, Carrie and Dylan. One of them is here with me. It is not Anne, it is not Carrie ...
Any numerical encoding necessarily introduces some ordering even where there is none, simply because numbers have order, whatever they may mean for us.
Even one-hot-encoding introduces order since $1$ is greater than $0$, right?
So any numerical encoding introduces order. One-hot might seem better, but if you ponder on the fact that it enlarges the ...