推理完后 GPU 没释放

[veasion]

执行 convert_video 函数，GPU 没有被释放 ：

import time
import torch
from model import MattingNetwork
from inference import convert_video

model = MattingNetwork('mobilenetv3').eval().cuda()
model.load_state_dict(torch.load('checkpoints/rvm_mobilenetv3.pth'))
convert_video(
    model,
    input_source='input.mp4',
    output_type='video',
    output_alpha='alpha.mp4',
    output_video_mbps=1,
    downsample_ratio=None,
    seq_chunk=12
)
time.sleep(10000)

[biubiue]

执行 convert_video 函数，GPU 没有被释放 ：

import time
import torch
from model import MattingNetwork
from inference import convert_video

model = MattingNetwork('mobilenetv3').eval().cuda()
model.load_state_dict(torch.load('checkpoints/rvm_mobilenetv3.pth'))
convert_video(
  model,
  input_source='input.mp4',
  output_type='video',
  output_alpha='alpha.mp4',
  output_video_mbps=1,
  downsample_ratio=None,
  seq_chunk=12
)
time.sleep(10000)

解决了么，兄弟

[veasion]

在推理代码后面手动清理一下gpu缓存

torch.cuda.empty_cache()