search

tarun005 / FLAVR

Code for FLAVR: A fast and efficient frame interpolation technique.
Apache License 2.0
463 stars 70 forks source link

How to use your model to predict the optical flow? #19

Closed jzzsc closed 3 years ago

jzzsc commented 3 years ago

Sorry to bother Is there any code for predicting the optical flow based on your model? I will appreciate it so much

tarun005 commented 3 years ago

We used the code available here https://github.com/ClementPinard/FlowNetPytorch . We first pretrain a model on Vimeo, drop the last prediction layer, attach another prediction layer for optical flow, and then train it on Sintel datasets. Results on 5-fold cross validation are provided in Table 6 in the paper.

jzzsc commented 3 years ago

We used the code available here https://github.com/ClementPinard/FlowNetPytorch . We first pretrain a model on Vimeo, drop the last prediction layer, attach another prediction layer for optical flow, and then train it on Sintel datasets. Results on 5-fold cross validation are provided in Table 6 in the paper.

Thank you so much for your reply. But I am still confused about how to use FLAVR as a faster method for optical flow prediction (Capter 6 in the paper). The input is 2CxHxWx3 figures, and the output is (k-1)xHxWx3 figures in the FLAVR architecture. Where does it calculate the optical flow (like EV-flow net)?

tarun005 commented 3 years ago

We drop the last output layer from FLAVR, and add a new prediction layer, which predicts the x and y components of the optical flow, and retrain it on a new flow datasets, like Sintel.

jzzsc commented 3 years ago

We drop the last output layer from FLAVR, and add a new prediction layer, which predicts the x and y components of the optical flow, and retrain it on a new flow datasets, like Sintel.

Thank you!