Thomas

Who has trained a model as good as the pretrained checkpoint model ?

2

There are too many hyper-parameters on train.lua and it is hard to train the model as good as the pretrained checkpoint model. The model I trained only achieve 62%~84% of...

Question about the FLOPs

3

I have two questions about FLOPs. (1) [here](https://github.com/microsoft/Swin-Transformer/blob/main/models/swin_transformer.py#L583) `` flops += self.num_features * self.patches_resolution[0] * self.patches_resolution[1] // (2 ** self.num_layers) `` I think it is the FLOPs for norm. We...

About the dataset

5

Hi @yl-1993, really an amazing work! I have some questions about the dataset. In https://github.com/yl-1993/learn-to-cluster/blob/master/DATASET.md 1. Is the feature of [ms1m_part1.tar.gz](https://drive.google.com/open?id=16WD4orcF9dqjNPLzST2U3maDh2cpzxAY) extracted with the [resnet50_part0_train.pth.tar](https://drive.google.com/open?id=1eKsh7x-RUIHhIJ1R9AlUjsJdsdbh2qim) ? 2. What is the...

Hi, what is the difference of center loss(old) and center loss(new)?

How to use skip_frames

1

I want to see the frames after top 3000 frames, I use ``` vr.skip_frames(3000) vr.seek(0) vr.next() ``` However, it is played from start. I use this code and it works,...

I cannot reproduce the 40.5 AP on yolox-s

2

I cannot reproduce the 40.5 AP with the command below ``` python -m yolox.tools.train -n yolox-s -d 8 -b 128 --fp16 -o --cache ``` I already use the batchsize=128 as...

Can you release the code about the genetic algorithm in the paper

Is this repo include the genetic algorithm on validate data?

1

Does this repo include the genetic algorithm on validate data? Can you share the code link?

1. 配置文件中 统一成了clip-vit-base-patch32，否则维度问题报错 2. train.py 中 stage index 依据 image_finetune 判断：stage = 1 if image_finetune else 2

This framework has reached 19.4 FPS with a single GeForce RTX 2080Ti GPU, using a batch size of 1. Can you tell us which version of torch and cuda this...