Guocode
Guocode
``` # create pytorch model from pytorch2keras import pytorch_to_keras from torch.autograd import Variable from models.mobilenet import PoseEstimationWithMobileNet model = PoseEstimationWithMobileNet(num_class=10) # create input tensor import numpy as np import torch...
``` num = torch.sum(torch.mul(predict, target)) + self.smooth den = torch.sum(predict.pow(self.p) + target.pow(self.p)) + self.smooth ``` The add of smooth is done on every index! It should be like this ```...
I draw the pose(yaw pitch roll) which is the second parameter of estimate_pose, it is right when yaw is small but it is wrong when yaw is large, here is...
In my opinion, if face image is cropped, s,t in pose param will change so that the model can match the image in pixel, but I didn't find the corresponding...
why policy network uses state[0] to be input but not the whole state? It is difficulty to understand that policy network predicts the whole network architecture by only the first...
Would you please update the evaluate result on 300W full
Can you tell more details about the imagenet+kinetics pretrain process? Is it to repeat the images of imagenet to form the input?
Hello, in your paper you use l1 ranking loss, but your code is l2 ranking loss, can you explain that. BTW, I have tried both and l2 outperforms l1 little.
dataset:something-somethingv2 slowfast only get 31% acc 3d-resnet-50 get 36% acc under the same settings. If anyone has do tests on any other dataset, you can publish your score here.
Hello, have you ever compared the different backbones of centernet, as my colleague said that the weak backbone will lead to a big reduction on result. I'm working on face...