ys-li

Results 16 comments of ys-li

lmdb dataset for REDS

Yes, you should build a custom dataset to support REDS in lmdb format. Besides, video data in lmdb format will be supported in V1.0

about testing of TDAN on SPMCS-30

Has the code been modified?

RuntimeError: Given groups=1, weight of size [64, 256, 1, 1], expected input[1, 128, 568, 320] to have 256 channels, but got 128 channels instead

We are sorry that FLAVR has not supported the VFI demo now.

RuntimeError: Given groups=1, weight of size [64, 256, 1, 1], expected input[1, 128, 568, 320] to have 256 channels, but got 128 channels instead

We are going to support it in #954

meaning of the name of pretrained VSR model

mid_channels=64, num_blocks=7, 8 GPUs, batch_size=1 in each of them. 600k items trained in REDS, x4.

meaning of the name of pretrained VSR model

We can modify a model to support x2 VSR or reproduction another x2 VSR model together.

runtime error

@Ashore-lz , please kindly post the error message

some problems about training

How about using 1 GPU or a smaller batch_size?

loading pretrained BasicVSR++

Maybe you can format the struct of the checkpoint file.

loading pretrained BasicVSR++

Or only load `state_dict` from the pre-trained checkpoint file. By `torch.load('FILE_PATH')['state_dict']`