RuntimeError: CUDA out of memory. about segment-and-track-anything HOT 5 CLOSED

同样地，我在运行demo.ipynb的“Generate Results for the Whole Video”这一部分代码时，也遇到了同样的问题：

Output exceeds the size limit. Open the full output data in a text editor---------------------------------------------------------------------------
RuntimeError Traceback (most recent call last)
Cell In[15], line 52
50 segtracker.add_reference(frame, pred_mask)
51 else:
---> 52 pred_mask = segtracker.track(frame,update_memory=True)
53 torch.cuda.empty_cache()
54 gc.collect()

File /media/alienware/新加卷1/Research/Segment-and-Track-Anything-main/SegTracker.py:75, in SegTracker.track(self, frame, update_memory)
67 def track(self,frame,update_memory=False):
68 '''
69 Track all known objects.
70 Arguments:
(...)
73 merged_mask: numpy array (h,w)
74 '''
---> 75 pred_mask = self.tracker.track(frame)
76 if update_memory:
77 self.tracker.update_memory(pred_mask)

File ~/anaconda3/envs/SAMTrack/lib/python3.9/site-packages/torch/autograd/grad_mode.py:28, in _DecoratorContextManager.call..decorate_context(*args, **kwargs)
25 @functools.wraps(func)
26 def decorate_context(*args, **kwargs):
27 with self.class():
...
579 dim=1).clamp(1e-5, 1 - 1e-5)
580 merged_logit = torch.logit(merged_prob)
582 return merged_logit

RuntimeError: CUDA out of memory. Tried to allocate 1.09 GiB (GPU 0; 7.78 GiB total capacity; 5.02 GiB already allocated; 182.88 MiB free; 5.21 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF

from segment-and-track-anything.

Hi, what pytorch version do you use? We recommend pytorch >= 1.10. If yes, please check how many objects are tracked when you use the notebook. If the object number is large, you may try using a smaller points_per_side, e.g. 8 and a larger sam_gap, e.g. 30 for your video to control the object number.

from segment-and-track-anything.

感谢，我试着调整了一下这两个参数，可以多处理一些帧，但是仍旧无法；处理整个视频。我想问一下你们使用的显卡配置，看看能不能向老师申请换个好点的，谢谢！

from segment-and-track-anything.

We use RTX 3090 24G, but according to our observation, 10G may be sufficient to run a video with upto 200 objects. There may be two solutions,
1, downscale your video to a smaller size, e.g. 640x320 and I guess it can be processed on a 8G GPU card. We are considering to add a parameter for users to control the size.
2, switch to a new enviroment with higher pytorch version and try again. The memory management mechanism may be different in different versions.

from segment-and-track-anything.

witch to a new e

非常感谢！

from segment-and-track-anything.