Crestina2001

How to specify the device

2

Hi, I am using the model for inference. I have several GPUs, and I would like to split the images for different GPUs to infer. How to move the restorer...

In the paper, you claim that: 3.5 Analysis and Discussion Taking the settings on the CULane [2] dataset as an example, we have H = 320, W = 1600, N_{row}...

With PyTorch>=0.5, the index of 0-dim tensor is invalid, so in train.py line 59 to 67, [0] should be replaced with .item(): e.g. loss.data[0] --> loss.data.item()

After epoch 5, the accuracy seems to fluctuate between 0.7 and 0.75.

An optimized version of compute_pairwise_distances in utils.py

1

The original function calculates pairwise distances by looping over all pairs of instances, which can be slow if the number of instances is large. We can use broadcasting and vectorization...

Best pdf extractor I have seen, but still not accurate enough

1

Thanks for your great work! But it still has some problems. I have a PDF, which is not scanned(you can select the words in the files). When using your method,...

After training for 20 epochs with default params, the validation results seem to be quite blurry.

4

 Could anyone offer a solution. Many thanks.

How to improve the performance

6

 I tested TAU on MMNist with the following config(the model part is copied from this repo): hid_S:16 hid_T:256 N_S:4 N_T:4 mlp_ratio:8 drop:0 drop_path:0 spatio_kernel_enc:3 spatio_kernel_dec:3 act_inplace:true T:10 width:64 height:64...

Can I adapt the model for video prediction(like moving mmnist)?

2

Thanks for the great work! Is it possible to do adapt the model for video prediction? And if so, what decoder model shall I use? Thanks for any suggestions!

Has anyone implemented a simplified version that only retain the RAG functions?(No need for the GUI, voice recognition, image recognition function, etc.) If not, how to do it? Any instructions?