LEI TAI

Why the value loss need to devide 2 in line 108 of a3c.py

1

v_loss += (v - R) ** 2 / 2 But the original paper just calculate the derivative of the (V-R)^2 right?

undefined references to `DisplayScreen::display_screen(MediaSource const&)

4

Hi, I use the caffe-dqn as the caffe, and when I do make, I got this. ``` CMakeFiles/dqn.dir/dqn_main.cpp.o: In function `PlayOneEpisode(ALEInterface&, dqn::DQN&, double, bool)': dqn_main.cpp:(.text+0xd14): undefined reference to `DisplayScreen::display_screen(MediaSource const&)'...

How to load resnet-101? (poor result based on resnet-50)

2

I trained the model with resnet-50 for 200k iterations. But the result is very poor. I wonder if we should use the resnet-101 as the original Mask-RCNN paper?

compared with the darknet yolo in real time with gpu.

https://github.com/HKUST-Aerial-Robotics/Btraj/blob/78156885dfab1c5dfeea96cf2ab7ecd772433381/third_party/fast_methods/gradientdescent/gradientdescent.hpp#L367 It seems that **ndims_** here is 3. And **d_** is declared with size of 2. However, **d_[2]** is initialized and used in folliwing codes. It works but is still...

From the official forward code, they normalize the input speed to one. https://github.com/carla-simulator/imitation-learning/blob/master/agents/imitation/imitation_learning.py#L141 And from the released dataset, it seemed that the speed is raw km/h (from 0 to 90)?...