jwchoi384/Gaussian_YOLOv3

count FP/TP

CuongNguyen218 opened this issue · 5 comments

CuongNguyen218 commented

Hi @jwchoi384 ,
I want to calculate TP/FP of KITTI dataset in your paper. How can I do it ?

jwchoi384 commented

@CuongNguyen218
I explained in detail in our paper. As mentioned our paper, we used bdd evaluation code.
When calculate TP and FP, we set detection threshold as 0.5 (test default threshold).

Gaussian_YOLOv3/examples/detector.c

Line 537 in 85d1d35

float thresh = 0.005;

In addition, IoU threshold is set as official evaluation threshold value of each dataset.

Gaussian_YOLOv3/bdd_evaluation/evaluate.py

Line 189 in 85d1d35

thresholds = [0.75]

CuongNguyen218 commented
jwchoi384 commented

You must print out the KITTI result as BDD format.
It is easy. Firstly, you just change eval=kitti to eval=bdd in cfg/KITTI_3cls.data.

Gaussian_YOLOv3/cfg/KITTI_3cls.data

Line 6 in 85d1d35

eval = kitti

Secondly, you must change the class id as KITTI classes. (default is BDD class id)

Gaussian_YOLOv3/examples/detector.c

Line 74 in 5d48f43

char *bdd_ids[] = {"car", "pedestrian", "cyclist"};

Finally, run the BDD evaluation code using bdd_evaluation/gt_KITTI_valid_for_bdd_eval_method.json.
This is the ground-truth file of KITTI validation set for using BDD evaluation method.
Before running the code, you must set the IOU threshold in bdd_evaluation/evaluate.py mentioned above (Car is 0.7, the others are 0.5).

CuongNguyen218 commented

@jwchoi384
I should change line https://github.com/ucbdrive/bdd-data/blob/master/bdd_data/evaluate.py#L189 from thresholds = [0.75] to thresholds = [0.7, 0.5, 0.5] right?
but when i change this line it throw an error.

CuongNguyen218 commented

how can i print the number of TP and FP compare to Gt box. Should I Print value of this line https://github.com/jwchoi384/Gaussian_YOLOv3/blob/master/bdd_evaluation/evaluate.py#L169, right ?