Hi,
model.eval()
with torch.no_grad():
for idx, (images, targets) in tqdm(enumerate(train_dl), total=len(train_dl)):
imgs, targs = model.transform(images, targets)
fv = model.backbone(imgs.tensors.to(device))["0"].cpu()
feature_vecs.append(fv)
While executing this code where model is a torchvision.models.detection.faster_rcnn
and images
and targets
are correct.
This loop abruptly stops at 45/1039 in the dataset. I have lots of memory. So why does it randomly terminate?