Dongwan Kim

**Current FFCV version**: v0.0.4 compiled from source. I have two identical FFCV `.beton` files (ImageNet train dataset): One in an NVME ssd drive, and the other in an HDD drive....

[Bug] Synchronization issue on GPU

14

I'm using the v0.0.4 version from this branch: https://github.com/libffcv/ffcv/tree/v0.0.4 There's a (possibly major) bug where two models will not receive the same inputs from the FFCV dataloader, unless `torch.cuda.synchronize` is...

Bug with transforms.Resize when used with transforms.ConvertImageDtype

11

### 🐛 Describe the bug Recent releases of Torchvision and the documentations that support it seem to suggest that we can use `io.read_image + transforms.ConvertImageDtype` instead of the traditional `PIL.Image.read_fn...

Incorrect accuracy calculation with DistributedDataParallel - examples/imagenet/main.py

6

The `accuracy()` function [here](https://github.com/pytorch/examples/blob/master/imagenet/main.py#L411) divides number of correct samples by the batch size. Then, the [top1 accuracy is updated](https://github.com/pytorch/examples/blob/master/imagenet/main.py#L340) with `AverageMeter`, and the `top1.avg` is returned. This is incorrect, since...

I've just encountered a weird issue where the PyTorch DataLoader will segfault in the following conditions: * FFCV and PyTorch dataloaders are both created * FFCV uses more than 2...

Question: How come the MaskDINO Encoder can't use half-precision (AMP) training?

1

So there's a decorator in the MaskDINOEncoder that disabled FP16 casting [here](https://github.com/IDEA-Research/detrex/blob/main/projects/maskdino/modeling/pixel_decoder/maskdino_encoder.py#L338). A few lines below a comment states that deformable-DETR does not support half precision (in Line 357). Do...