Obtaining .npy files

[florallam]

Hello! Thank you for sharing your repository, I do have some questions about evaluating the model using the .npy file. As per a previous issue, the reply mentioned that the test.txt should be created.

I realised that the image pairs cannot be generated at random, as line 108 of the dataloader will find the corresponding .npy file involved. How then may I generate the .npy files? Thank you!

[captainfffsama]

this work dataset seems follow DeepHomography, in deephomography paper,they said :

For each pair, we manually marked 6 ∼ 8 equally distributed matching points for the purpose of quantitative comparisons, as illustrated in the rightmost column of Fig. 5.

So I create a simple tools to label the homo matrix of image pair,may be u can use this tools to label the homo matrix then convert json label to npy.
https://github.com/captainfffsama/LabelHomography