What do all of these parameters from training YOLOv4 mean?
(next mAP calculation at 1300 iterations)
Last accuracy mAP@0.5 = 63.16 %, best = 68.55 %
1249: 26.351213, 24.018257 avg loss, 0.001000 rate, 2.983998 seconds, 39968 images, 10.505599 hours left Loaded: 0.000068 seconds
(next mAP calculation at 1300 iterations) Last accuracy mAP@0.5 = 63.16 %, best = 68.55 %
1250: 13.904115, 23.006844 avg loss, 0.001000 rate, 4.093653 seconds, 40000 images, 10.456502 hours left Resizing, random_coef = 1.40
Here's what the parameters mean.
For your given example:
(next mAP calculation at 1300 iterations) Last accuracy mAP@0.5 = 63.16 %, best = 68.55 %
1250: 13.904115, 23.006844 avg loss, 0.001000 rate, 4.093653 seconds, 40000 images, 10.456502 hours left Resizing, random_coef = 1.40
1250 --> iteration
Last accuracy mAP@0.5 --> Last mean average precision (mAP) at 50% IoU threshold. mAP is calculated every 100th iteration. So, in the example, it's the mAP from iteration = 1200
best --> highest mAP so far
13.904115 --> total loss
23.006844 avg loss--> average loss, this is the thing you should care about for being low in training
0.001000 rate --> learning rate
4.093653 seconds --> total time spent to process the batch
40000 images --> total amount of images used during training so far (iteration*batch = 1250 * 32)
10.456502 hours left --> estimated time remaining for finishing up to the max_batches in your config file
Resizing, random_coef = 1.40 --> Confirming that your dataset is being randomly resized every 10 iterations from 1/1.4 to 1.4 (in this iteration, it's 1.40)
References: https://github.com/AlexeyAB/darknet/blob/master/src/detector.c https://github.com/AlexeyAB/darknet/wiki/CFG-Parameters-in-the-different-layers