Damien Menigaux
Damien Menigaux
@peterbell10 could you please advise me ? In all honesty Idk what I'm doing. I understand that the strides of the grad output are making this test fail, but I...
Hi. Thank you for your interest. I haven't produced one yet but I'll keep you updated if / when I do. If someone has one to share, he / she...
Me neither on Kinetics. ALthough I'm starting from scratch. I'm in the process of pretraining it with boring imagenet videos now and I'll see how good it performs afterwards.
Hi ! I've modified it to a great extent. You should try it again. I've made Imagenet pretraining work
I've managed to train it on kinetics, but I'm not using the parameters from the paper. So I'm not sure It's worth putting the weights here. I'll try and reproduce...
Hi. Unfortunately I'm unfamiliar with any FLOPS estimation tool. But as I am not aware of all the hyperparameters used by the team that wrote the paper, it's probable that...
Lateral connections is the heart of it. Might as well not have released any code at all....
I'm in the process of training it. I'll keep you updated.
The results on kinetics are competitive, but I'm not able right now to provide the petrained weights that you guys will want, because I'm using a configuration for my own...
> @SidHard I find it's essential to pretrain for 3DCNN on imagenet, is your test of I3D and slowfastnet both trained from scratch? The paper exhibits results with random init,...