search

JoyHuYY1412 / LST_LVIS

MIT License
47 stars 6 forks source link

What do variable "id_step1", "nStep", "nBase" refer to in file balanced_replay.ipynb? #1

Open isunLt opened 3 years ago

isunLt commented 3 years ago

Hi @JoyHuYY1412 , thanks for sharing your codes here. You've proposed a novel paradigm for the LVIS task. I've met a problem when trying to run your code. Some variable like "id_step", "nStep", "nBase" seem undefined in the file balanced_replay.ipynb . I gauss "nStep" may refer to "step_size", and "nBase" refer to "base_size" according to context. So what do these variables actually refer to?

JoyHuYY1412 commented 3 years ago

Hi @JoyHuYY1412 , thanks for sharing your codes here. You've proposed a novel paradigm for the LVIS task. I've met a problem when trying to run your code. Some variable like "id_step", "nStep", "nBase" seem undefined in the file _balancedreplay.ipynb . I gauss "nStep" may refer to "step_size", and "nBase" refer to "base_size" according to context. So what do these variables actually refer to?

So sorry for some omission. You are right, nStep is step_size, e.g. 160; nBase is size of the base classes, e.g, 270. id_step is the class id list in the current step. Hope this can help!

isunLt commented 3 years ago

Hi @JoyHuYY1412 , thanks for sharing your codes here. You've proposed a novel paradigm for the LVIS task. I've met a problem when trying to run your code. Some variable like "id_step", "nStep", "nBase" seem undefined in the file _balancedreplay.ipynb . I gauss "nStep" may refer to "step_size", and "nBase" refer to "base_size" according to context. So what do these variables actually refer to?

So sorry for some omission. You are right, nStep is step_size, e.g. 160; nBase is size of the base classes, e.g, 270. id_step is the class id list in the current step. Hope this can help!

Thanks for your reply! So, you mean that id_step contains the replayed class id of the last step and the class id of the current step?

JoyHuYY1412 commented 3 years ago

Hi @JoyHuYY1412 , thanks for sharing your codes here. You've proposed a novel paradigm for the LVIS task. I've met a problem when trying to run your code. Some variable like "id_step", "nStep", "nBase" seem undefined in the file _balancedreplay.ipynb . I gauss "nStep" may refer to "step_size", and "nBase" refer to "base_size" according to context. So what do these variables actually refer to?

So sorry for some omission. You are right, nStep is step_size, e.g. 160; nBase is size of the base classes, e.g, 270. id_step is the class id list in the current step. Hope this can help!

Thanks for your reply! So, you mean that _idstep contains the replayed class id of the last step and the class id of the current step?

id_step should contain the classes you need to train in this step, like 270+160*n in step n.