Jensen (Jinghao) Zhou

Hi, @ChunyuanLI. I have been trying to download the checkpoint to load the pre-training args, but the download speed was extremely slow and the download often failed halfway. Could you...

Hi @ChunyuanLI @jwyang, I'm also wondering about the linear probe accuracy of DINO/EsViT with Swin-L. Have you ever run any related experiments?

Hi @merlinarer, I'm no longer able to access this log. Are you able to reproduce the results following the [instructions](https://github.com/bytedance/ibot/tree/main/evaluation#semi-supervised-classification-on-imagenet)?

Close it for now. Feel free to re-open.

Hi @Joker316701882 , Did you use the default fine-tuning recipe for supervised Swin-T? If yes, you could probably try to tweak the fine-tuning learning rate, layer decay rate, etc. Setting...

> By keeping your default parameters, without fine-tune, I got 0.6466 on OTB2013 and 0.6206 on OTB2015 (guess you are using py2 so I also tested using py2). So is...

@LCAR979 I changed to python 2.7 but got 0.6336/0.6014. Could you offer more details about the package versions?

> @maniac-py As is described in [DaSiamRPN paper](http://openaccess.thecvf.com/content_ECCV_2018/papers/Zheng_Zhu_Distractor-aware_Siamese_Networks_ECCV_2018_paper.pdf), the model is trained on ImageNetVID, YouTube-BB, COCO and ImageNet Detection datasets. > > The model in this repo is directly translated...

Hi @ShnitzelKiller, I've encountered the same problem. Have you figured out how to tackle this?