The predicted location seems to be very different from the groundtruth...
hnyz979 opened this issue · comments
Hi, Dr. Ning
I am now trying your first step, namely, simply reproducing the results with the pre-trained model with the "ROLO_network_test_all.py" code. I have downloaded "model_demo.ckpt" and ran the codes. The codes seems to be working, but I am a little confused about the results:
(The following are the results of the first and second iteration, and I uncommented the codes to print out "batch_ys" and "pred_location", which seems to be the ground truth and the predicted location. However, the results are as follows:)
('Batch_ys: ', array([[ 0.51145833, 0.64296875, 0.19791667, 0.5078125 ]]))
('ROLO Pred: ', array([[-0.07923603, -0.10261807, 0.09152682, 0.03067834]], dtype=float32))
('len(pred) = ', 1)
('ROLO Pred in pixel: ', -38.033294677734375, -65.675563812255859, 43.932874202728271, 19.634140729904175)
('correct_prediction int: ', array([[0, 0, 0, 0]]))
Iter 0, Minibatch Loss= 28.594887
('Batch_ys: ', array([[ 0.51354167, 0.64296875, 0.19791667, 0.5078125 ]]))
('ROLO Pred: ', array([[-0.07972586, -0.10288554, 0.09197099, 0.03046498]], dtype=float32))
('len(pred) = ', 1)
('ROLO Pred in pixel: ', -38.268413543701172, -65.846743583679199, 44.146074056625366, 19.497586488723755)
('correct_prediction int: ', array([[0, 0, 0, 0]]))
Iter 1, Minibatch Loss= 28.683752
It seems that, the Batch_ys(groundtruth) are very different from the ROLO Pred(ROLO predictions). Could you kindly tell me where did I do wrong? Thank you very much! @Guanghan
@hnyz979 我也遇到和你同样的问题
实际上，在经过一段时间的尝试后，我无法重复论文中所描述的预测结果，无论是实验1还是实验2。我已经转向其他work
@lihaixiang 你试过YOLO没？那个能跑起来吗？
@hnyz979 可以
