Training issues

[lydcv]

I'm sorry to disturb your rest. Could you please share the calculation code for FLOPS, Param, and FPS used for meta-learning method FPD? I haven't completed it yet, and I hope you can help me despite your busy schedule. Thank you very much!

[wangchen1801]

from ptflops import get_model_complexity_info
from thop import profile
for idx, data_batch in enumerate(data_loader):
    macs, params = get_model_complexity_info(model, (1, 3, 800, 1333), as_strings=True,
                                             print_per_layer_stat=False, verbose=True)
    print('{:<30}  {:<8}'.format('Computational complexity: ', macs))
    print('{:<30}  {:<8}'.format('Number of parameters: ', params))

    # flops, params = profile(model.cuda(), inputs=data_batch['img'][0].cuda())
    # print('FLOPs = ' + str(flops / 1000 ** 3) + 'G')
    # print('Params = ' + str(params / 1000 ** 2) + 'M')

you can try 'profile' and 'get_model_complexity_info' to calculate FLOPS and Params.

[lydcv]

from ptflops import get_model_complexity_info
from thop import profile
for idx, data_batch in enumerate(data_loader):
    macs, params = get_model_complexity_info(model, (1, 3, 800, 1333), as_strings=True,
                                             print_per_layer_stat=False, verbose=True)
    print('{:<30}  {:<8}'.format('Computational complexity: ', macs))
    print('{:<30}  {:<8}'.format('Number of parameters: ', params))

    # flops, params = profile(model.cuda(), inputs=data_batch['img'][0].cuda())
    # print('FLOPs = ' + str(flops / 1000 ** 3) + 'G')
    # print('Params = ' + str(params / 1000 ** 2) + 'M')

you can try 'profile' and 'get_model_complexity_info' to calculate FLOPS and Params.

May I ask if you would like to directly place it in test. py and select the base training weight of 20000pth or a fine tuned weight such as the 1-shot weight of 800.pth for testing and printing? Thank you again for your help despite your busy schedule, and I wish you a happy new year!!

[wangchen1801]

Yes, you can either put it in test.py or manually load an image. Happy new year too!

[lydcv]

Yes, you can either put it in test.py or manually load an image. Happy new year too!
I'm sorry, I still got an error. I placed it at the end of test. py, but when executing Python test. py configs/fpd/voc/split1/fpd_r101_c4_2xb4_oc-split1_base training. py work-dirs/split1_base/latest.tpth -- eval mAP, the error still appears. Thank you for taking the time to help again! The error is as follows:
OrderedDict([('AP50', 0.907), ('mAP', 0.9074152708053589)])
Warning: module Bottleneck is treated as a zero-op.
Warning: module ResLayer is treated as a zero-op.
Warning: module ResNetWithMetaConv is treated as a zero-op.
Warning: module Sigmoid is treated as a zero-op.
Warning: module MetaRCNNResLayer is treated as a zero-op.
Warning: module RoIAlign is treated as a zero-op.
Warning: module SingleRoIExtractor is treated as a zero-op.
Warning: module CrossEntropyLoss is treated as a zero-op.
Warning: module SmoothL1Loss is treated as a zero-op.
Warning: module MetaBBoxHead is treated as a zero-op.
Warning: module Embedding is treated as a zero-op.
Warning: module PrototypesDistillation is treated as a zero-op.
Warning: module PrototypesAssignment is treated as a zero-op.
Warning: module L1Loss is treated as a zero-op.
Warning: module RPNHead is treated as a zero-op.
Warning: module FPDRoIHead is treated as a zero-op.
Warning: module FPD is treated as a zero-op.
Warning: module MMDataParallel is treated as a zero-op.
Traceback (most recent call last):
File "test.py", line 265, in
main()
File "test.py", line 255, in main
macs, params = get_model_complexity_info(model, (1, 3, 800, 1333), as_strings=True,
File "/root/miniconda3/lib/python3.8/site-packages/ptflops/flops_counter.py", line 43, in get_model_complexity_info
_ = flops_model(batch)
File "/root/miniconda3/lib/python3.8/site-packages/torch/nn/modules/module.py", line 889, in _call_impl
result = self.forward(*input, **kwargs)
File "/root/miniconda3/lib/python3.8/site-packages/mmcv/parallel/data_parallel.py", line 42, in forward
return super().forward(*inputs, **kwargs)
File "/root/miniconda3/lib/python3.8/site-packages/torch/nn/parallel/data_parallel.py", line 165, in forward
return self.module(*inputs[0], **kwargs[0])
File "/root/miniconda3/lib/python3.8/site-packages/torch/nn/modules/module.py", line 889, in _call_impl
result = self.forward(*input, **kwargs)
File "/root/miniconda3/lib/python3.8/site-packages/mmcv/runner/fp16_utils.py", line 98, in new_func
return old_func(*args, **kwargs)
File "/root/autodl-tmp/FPD/fpd/query_support.py", line 179, in forward
return self.forward_train(query_data, support_data, **kwargs)
File "/root/autodl-tmp/FPD/fpd/query_support.py", line 269, in forward_train
query_img = query_data['img']
TypeError: new(): invalid data type 'str'

[lydcv]

May I ask if you have added forward_dummy to fpd_detector.py to handle it? Could you share the complete code process? I'm really sorry to bother you all the time. Thank you for taking the time to help me!