search

keyu-tian / SparK

[ICLR'23 Spotlight🔥] The first successful BERT/MAE-style pretraining on any convolutional network; Pytorch impl. of "Designing BERT for Convolutional Networks: Sparse and Hierarchical Masked Modeling"
https://arxiv.org/abs/2301.03580
MIT License
1.41k stars 82 forks source link

Spark.forward #61

Closed ricklisz closed 10 months ago

ricklisz commented 10 months ago

Dear author,

I wonder what is the use of SparK.forward in line 156 in main.py from pretrain? Is that necessary to call the forward function? I believe the defined model is actually model_without_ddp.

Similar issue in pretrain/viz_reconstruction, where the function show calls another spark.forward.

Looking forward to your response!

keyu-tian commented 10 months ago

@ricklisz those forward are useless and can be removed. I wrote them just for more hints from my IDE.