Hi,
Time is running out and I am not sure I will be able to submit. Here’s an overview of my pipeline, and where I am now:
STEP 1. Take each video and create new videos cropped to 384x384 centered in the zone of interest (where the fish action is happening) and rotating them so all fish / ruler is horizontal. DONE.
I’ve manually verified all test 667 videos and all look OK, e.g. one of the boat scenarios where the fish is on the edge of the screen ends up like this:
STEP 2. Take all videos created in step 1) and build frame by frame predictions of length, fish species including a new species_no_fish when there’s no fish. DONE.
Here’s an example of the generated predictions for the training data:
0,00WK7DR6FyPZ5u3A,0,157.63180541992188,1.7653231099146183e-09,1.0,3.6601498720756354e-08,3.1204638162307674e-07,2.3616204103404925e-08,5.9193606460894443e-08,1.3712766033791013e-08,6.261254969358587e-12
1,00WK7DR6FyPZ5u3A,1,162.67236328125,1.2406671523468304e-10,1.0,2.6489321847122937e-09,1.1546971734333056e-07,9.599543382421416e-10,4.391315311380595e-09,5.5750559724288e-10,1.2707434930703254e-13
2,00WK7DR6FyPZ5u3A,2,155.6374969482422,8.897231285054374e-10,0.9999971389770508,1.4144226270218496e-06,1.3051650284978678e-06,1.1586666737173346e-08,8.389072547743126e-08,3.659388880805636e-08,9.07268617872381e-12
3,00WK7DR6FyPZ5u3A,3,171.65701293945312,5.965576366229186e-10,0.9999971389770508,3.2907976788010274e-07,2.809475745380041e-06,5.564644856015377e-10,3.567773987356304e-08,7.221697462256316e-09,2.0023773750210694e-09
4,00WK7DR6FyPZ5u3A,4,140.32904052734375,1.1474267716526931e-11,8.804387441330164e-09,3.527556580174007e-10,2.4278079546746767e-09,2.6827713228011474e-12,2.4495312445083073e-08,8.0641066740883e-12,1.0
5,00WK7DR6FyPZ5u3A,5,142.9277801513672,3.1375051445792224e-09,9.983174464878175e-08,2.7918925837866482e-08,1.430677087910226e-07,2.2644339736643815e-09,1.3154048247088213e-05,3.765904033059542e-09,0.9999866485595703
6,00WK7DR6FyPZ5u3A,6,132.2730255126953,4.726588564984979e-11,1.3540606857986859e-08,2.1345873957301364e-09,1.7951052200260165e-08,1.2603748773820644e-11,4.9515310962533476e-08,1.5555539495393234e-11,1.0
7,00WK7DR6FyPZ5u3A,7,132.52487182617188,8.98270624549724e-11,2.9316490568476183e-08,1.4304294637668136e-08,2.306131676732548e-07,8.337074780540021e-11,1.8565401660453063e-07,1.1116094156271572e-10,1.0
8,00WK7DR6FyPZ5u3A,8,134.51568603515625,3.730782696664825e-11,1.0075987155744315e-08,1.2921342884553155e-09,3.362889700042615e-08,1.736233032345602e-11,9.184835647602085e-08,1.3911538587763062e-11,1.0
9,00WK7DR6FyPZ5u3A,9,129.33639526367188,3.3274737132327203e-11,2.59468269092622e-08,1.3075700522335865e-09,1.849521602537152e-08,1.1054876632166089e-11,6.560475895867057e-08,2.9239603793751456e-11,1.0
10,00WK7DR6FyPZ5u3A,10,134.68255615234375,2.8945401631119694e-11,4.223948302239933e-09,1.1841199132334168e-09,6.319635037499438e-09,4.090854825028467e-12,3.452311148066656e-08,1.3991316785699759e-11,1.0```
STEP 3. Take the csv predictions generated in Step 2) and predict the `fish_number` column. I'm building a RNN to do this but Im stuck (I decided to use this competition to learn Pytorch so have been progressing slower than expected). I understand non-RNN approaches for `fish_number` prediction may work.
Basically if you are interested in moving forward with STEP 3 and would like to merge please contact me.
-Andres