when encoding pytorch generated frames with PyNvVideoCodec, frame order is incorrect. 1.0.2 works fine but 2.0.0, 2.0.1 and 2.0.2 won’t work correctly
to replicate:
construct a pytorch tensor sequence in yuv420 named images
for i in range (N):
input_tensor = images[i,:,:,:]
torch.cuda.synchronize()
bitstream = self.encoder.Encode(input_tensor)
bitstream = bytearray(bitstream)
Will observe that encoded video have permuted (stuttering) frame. The permuting window is around 32 frames
NVIDIA-SMI 570.86.10 Driver Version: 570.86.10 CUDA Version: 12.8
GPU: 5090
Codec used: hevc