How to render image using custom camera intrinsics

[LuckyOne09]

Hello, I notice screenshot_transforms in run.py and I can use it to render image given custom camera poses.
However, how should I give camera intrinsics additionally?
I have went through python_api.cu and tried codes like these:

testbed.fov_axis = 0
for idx in args.screenshot_frames:
          f = ref_transforms["frames"][int(idx)]
          cam_matrix = f["transform_matrix"]
          testbed.fov_xy = [
	          f["camera_angle_x"] * 180 / np.pi, 
	          f["camera_angle_y"] * 180 / np.pi
          ]
          testbed.set_nerf_camera_matrix(np.matrix(cam_matrix)[:-1,:])
          image = testbed.render(args.width or int(ref_transforms["w"]), args.height or int(ref_transforms["h"]), args.screenshot_spp, True)

But the results are not right. Does anyone could give some guidelines? Very thanks