SSA and MSA code

[yuanpf123]

作者你好，请问single-source self-paced curriculum adapting (SSA) and multi-source self-paced curriculum adapting (MSA) algorithms这两个算法发布了吗，我没有找到相关的代码

[linhuixiao]

Hi, the SSA and MSA are algorithmic ideas for filtering and training pseudo-labels. Due to the complexity of their implementation, it is not convenient to organize them into a modular function for direct calling. Therefore, we do not provide the code for SSA and MSA. However, we do provide the code (function evaluate_for_filtering()) for evaluating the Reliability of samples in the file https://github.com/linhuixiao/CLIP-VG/blob/master/engine.py#L131.

[KevinfromTJ]

Hello, thank you for this great work!
I would like to know whether the curriculum scheduler used in this work is off-the-shelf from a 3rd library or needs significant modification or elaborate self-design ?
And I have some questions about the SSA and MSA process. How much time does it cost to finish a complete SSA or MSA process? I notice the training FPS in the paper is really high but it seems to have nothing to do with the SSA or MSA? Do they cost much time on the iterative process to train, eval and select data-subsets?