Hello author, I have trained three training strategies based on your github intro steps and obtained ttnet_3rd_phase_epoch_30pth. However, both the test and demo results are very poor. The PCE is only over 50 and the SPCE is over 70, and the output frames of the demo are very blurry. The frames' properties are 1920 * 1080, but the actual estimate is only 320 * 128, and it cannot correctly mark the position of the ball.
Hello author, I have trained three training strategies based on your github intro steps and obtained ttnet_3rd_phase_epoch_30pth. However, both the test and demo results are very poor. The PCE is only over 50 and the SPCE is over 70, and the output frames of the demo are very blurry. The frames' properties are 1920 * 1080, but the actual estimate is only 320 * 128, and it cannot correctly mark the position of the ball.