[Reproducing quantitative numbers of AOT and DeAOT] about aot-benchmark HOT 7 CLOSED

@z-x-yang Your fast response is highly appreciated.

from aot-benchmark.

How did you load the checkpoint for the model? Please make sure you are using the right ckeckpoint file for AOTL, and it is correctly loaded. You can check the output log of evaluation.

from aot-benchmark.

Hello, @yoxu515 .
It is such an honor to be able to access your code.
I tried the inference through the pretrained weights you provided, but it doesn't work well. I don't know, do I have to do the inference with the weights printed out by training myself? Or is it possible to use the pretrained weights that you provided?

Hello, @Amshaker
Looking at the issues you left above, I think you succeeded, can you share how you did the inference? I really need your help because the situation is urgent.

from aot-benchmark.

Also, I would appreciate it if you could tell me how to solve the error "No module named 'spatial_correlation_sampler' ".

from aot-benchmark.

Hi @cekkec
You can directly infer with the pretrained weights we provided in the model zoo.
Did you get similar results as Amshaker? If so, please tell us the model you use and what changes have you made on the code. We will have a check on this.
You may check the output log to see whether the model is loaded correctly.
As for the 'spatial_correlation_sampler', you can follow the instruction in readme to install it.

from aot-benchmark.

@yoxu515
Thank you very much for your quick response.
I proceeded with the code below.

dv_data = '/Users/cekke/Desktop/workspace/Dataset_VOS/DAVIS'
pth_load = '/Users/cekke/Desktop/workspace/ExRE/aot-benchmark/result/default_AOTT/PRE_YTB_DAV/ckpt/ckpt/SwinB_AOTL_PRE_YTB_DAV.pth'
def main():
    import argparse
    parser = argparse.ArgumentParser(description="Eval VOS")
    parser.add_argument('--exp_name', type=str, default='default') #---

    parser.add_argument('--stage', type=str, default='pre_ytb_dav') #---
    parser.add_argument('--model', type=str, default='SwinB_AOTL') #---
    parser.add_argument('--lstt_num', type=int, default=-1)
    parser.add_argument('--lt_gap', type=int, default=-1)
    parser.add_argument('--st_skip', type=int, default=-1)
    parser.add_argument('--max_id_num', type=int, default='-1')

    parser.add_argument('--gpu_id', type=int, default=0)
    parser.add_argument('--gpu_num', type=int, default=1) #---
    
    parser.add_argument('--ckpt_path', type=str, default=pth_load)
    parser.add_argument('--ckpt_step', type=int, default=-1)

    parser.add_argument('--dataset', type=str, default=dv_data) #---
    parser.add_argument('--split', type=str, default='val') #---

    parser.add_argument('--ema', action='store_true')
    parser.set_defaults(ema=False)

    parser.add_argument('--flip', action='store_true')
    parser.set_defaults(flip=False)
    parser.add_argument('--ms', nargs='+', type=float, default=[1.])

    parser.add_argument('--max_resolution', type=float, default=480 * 1.3)

    parser.add_argument('--amp', action='store_true')
    parser.set_defaults(amp=False)

Thank you so much for taking the time out of your busy schedule.

from aot-benchmark.

Hi, @cekkec
Can you send me your output log file during evaluation? For example, you can redirect the output to a log file by adding >> after your command, like "python eval.py [some args...] >> log.txt". My email is [email protected]. BTW, have you got the J&F score of your result?

from aot-benchmark.