open-mmlab/mmfewshot

OpenMMLab FewShot Learning Toolbox and Benchmark

 

OpenMMLab website ^HOT      OpenMMLab platform ^{TRY IT OUT}

 

📘Documentation | 🛠️Installation | 👀Model Zoo | 🆕Update News | 🤔Reporting Issues

English | 简体中文

Introduction

mmfewshot is an open source few shot learning toolbox based on PyTorch. It is a part of the OpenMMLab project.

The master branch works with PyTorch 1.5+. The compatibility to earlier versions of PyTorch is not fully tested.

Major features

• Support multiple tasks in Few Shot Learning

  MMFewShot provides unified implementation and evaluation of few shot classification and detection.

• Modular Design

  We decompose the few shot learning framework into different components, which makes it much easy and flexible to build a new model by combining different modules.

• Strong baseline and State of the art

  The toolbox provides strong baselines and state-of-the-art methods in few shot classification and detection.

What's New

v0.1.0 was released in 24/11/2021. Please refer to changelog.md for details and release history.

Installation & Dataset Preparation

MMFewShot depends on PyTorch and MMCV. Please refer to install.md for installation of MMFewShot and data preparation for dataset preparation.

Getting Started

If you are new of few shot learning, you can start with learn the basics. If you are familiar with it, check out getting_started.md for the basic usage of mmfewshot.

Refer to the below tutorials to dive deeper:

• Few Shot Classification

  • Overview
  • Config
  • Customize Dataset
  • Customize Model
  • Customize Runtime

• Few Shot Detection

  • Overview
  • Config
  • Customize Dataset
  • Customize Model
  • Customize Runtime

Benchmark and model zoo

Results and models are available in the model zoo. Supported algorithms:

Classification

• Baseline (ICLR'2019)
• Baseline++ (ICLR'2019)
• NegMargin (ECCV'2020)
• MatchingNet (NeurIPS'2016)
• ProtoNet (NeurIPS'2017)
• RelationNet (CVPR'2018)
• MetaBaseline (ICCV'2021)
• MAML (ICML'2017)

Detection

• TFA (ICML'2020)
• FSCE (CVPR'2021)
• AttentionRPN (CVPR'2020)
• MetaRCNN (ICCV'2019)
• FSDetView (ECCV'2020)
• MPSR (ECCV'2020)

Contributing

We appreciate all contributions to improve mmfewshot. Please refer to CONTRIBUTING.md in MMFewShot for the contributing guideline.

Acknowledgement

mmfewshot is an open source project that is contributed by researchers and engineers from various colleges and companies. We appreciate all the contributors who implement their methods or add new features, as well as users who give valuable feedbacks. We wish that the toolbox and benchmark could serve the growing research community by providing a flexible toolkit to reimplement existing methods and develop their own new methods.

Citation

If you find this project useful in your research, please consider cite:

@misc{mmfewshot2021,
    title={OpenMMLab Few Shot Learning Toolbox and Benchmark},
    author={mmfewshot Contributors},
    howpublished = {\url{https://github.com/open-mmlab/mmfewshot}},
    year={2021}
}

License

This project is released under the Apache 2.0 license.

Projects in OpenMMLab

• MMCV: OpenMMLab foundational library for computer vision.
• MIM: MIM installs OpenMMLab packages.
• MMClassification: OpenMMLab image classification toolbox and benchmark.
• MMDetection: OpenMMLab detection toolbox and benchmark.
• MMDetection3D: OpenMMLab's next-generation platform for general 3D object detection.
• MMRotate: OpenMMLab rotated object detection toolbox and benchmark.
• MMSegmentation: OpenMMLab semantic segmentation toolbox and benchmark.
• MMOCR: OpenMMLab text detection, recognition and understanding toolbox.
• MMPose: OpenMMLab pose estimation toolbox and benchmark.
• MMHuman3D: OpenMMLab 3D human parametric model toolbox and benchmark.
• MMSelfSup: OpenMMLab self-supervised learning Toolbox and Benchmark.
• MMRazor: OpenMMLab Model Compression Toolbox and Benchmark.
• MMFewShot: OpenMMLab FewShot Learning Toolbox and Benchmark.
• MMAction2: OpenMMLab's next-generation action understanding toolbox and benchmark.
• MMTracking: OpenMMLab video perception toolbox and benchmark.
• MMFlow: OpenMMLab optical flow toolbox and benchmark.
• MMEditing: OpenMMLab image and video editing toolbox.
• MMGeneration: OpenMMLab Generative Model toolbox and benchmark.
• MMDeploy: OpenMMlab deep learning model deployment toolset.

mmfewshot's Issues

--resume-from ignores last best accuracy

When resuming training again, the best accuracy count starts from scratch, i.e, it ignores the previous best accuracy.

About the limited gpus for few-shot detection?

How many gpus will be used for the few-shot detection,whether it support 1 gpu?

unexpected key in source state_dict

in my log, it will be see: unexpected key in source state_dict
if i set pretrained = 'open-mmlab://detectron2/resnet50_caffe' in configs/detection/base/models/faster_rcnn_r50_caffe_c4.py, the log print "unexpected key in source state_dict: conv1.weight, ......"
if i set pretrained = 'iter_20000.pth', this is the saved model, the log print "unexpected key in source state_dict: backbone.conv1.weight, ......"
does it mean I didn't load any parameters? how to solve this problem?

test query set would have some samples from test support set

I reload your model from metafile.yml and want to test the performance of MMfewshot in my dataset.
I found that when I input dataset with different numbers of images like 50, 100 ,200 ,400 , the more images I input , the less acc I got. And I found that datawraper would load same images into support set and query set . The less images I input ,the higher acc is.
There is my result.

<style> </style>

model=meta-baseline
	800 eposide
	num querys 15
	5 way 5shot		5 way 10shot		5 way 15shot		5 way 20shot
input_pics	acc	acc std	acc	acc std	acc	acc std	acc	acc std
50	69.7	0.46	75.91	0.43	78.05	0.4	79.6	0.36
100	63.77	0.47	69.33	0.41	71.22	0.38	72.4	0.36
200	60.63	0.43	66.48	0.41	68.59	0.4	70.24	0.4
400	59.76	0.43	65.44	0.4	67.5	0.38	68.93	0.38

Few-shot instead of one-shot in demo inference

Currently, the demo script (classification) takes only one sample in the support set. It uses the process_support_images() method to forward the support set. How to modify this in order to allow for more than one sample in the support set?

One idea could be to place another set of support images in a different folder and then forward that as well. Then the model.before_forward_support() method can be modified if it resets the features. For e.g. for meta_baseline_head, it is resetting saved features.

Then (again for meta_baseline), meta_baseline_head.before_forward_query would also have to be modified since it is replacing the self.mean_support_feats with the mean of the new support set.

Would these two changes in this case be enough to adapt for a few-shot instead of a one-shot inference?

For VOC dataset, How to get AP and AP75？

For VOC dataset, How to get AP and AP75？

Is it possible to provide a 5-shot classification demo program?

I can successfully run the 1-shot demo program, but I don't know how to inference about the 5-shot model. "DEMO"

How to prepare my own datasete like coco and voc format to do fewshot training?

Looking forward to your reply. Thanks!

Loss does not decrease

I used the coco dataset and the default parameter configuration in 'base_train', but we have reached the 2w iteration, and the loss has not decreased. Why?

Is there a plan to support Windows platform ?

I see Linux (Windows is not officially supported)
Is there a plan to support Windows platform? Thanks.

add support for facebookresearch/detic

Its a high performing zero shot object detector with changeable vocabulary: https://github.com/facebookresearch/Detic

bad performance of Attention RPN in base training

When i train the attention rpn model in the base training stage, the performance is bad and far away from that of other detectors (meta rcnn), what may be the reason? By the way, is the way always set as 2, and how is the shot is set, it is inconsistent with the shot num in config files (such as shot=9 in 10shot config yaml file).
Thanks

Error when plotting image recognition results

First of all kudos for the great repo!

There is a small problem with plotting the results here. When running the image demo demo_metric_classifier_1shot_inference for the sample classes the following error occurs:

Traceback (most recent call last):
  File "demo/demo_metric_classifier_1shot_inference.py", line 51, in <module>
    main()
  File "demo/demo_metric_classifier_1shot_inference.py", line 47, in main
    show_result_pyplot(args.image, result)
  File "mmfewshot/mmfewshot/classification/apis/inference.py", line 150, in show_result_pyplot
    img = imshow_infos(
  File "mmfewshot/lib/python3.8/site-packages/mmcls/core/visualization/image.py", line 316, in imshow_infos
    img = manager.put_img_infos(
  File "/home/user/.pyenv/versions/mmfewshot/lib/python3.8/site-packages/mmcls/core/visualization/image.py", line 277, in put_img_infos
    self.wait_continue(timeout=wait_time)
  File "/home/user/.pyenv/versions/mmfewshot/lib/python3.8/site-packages/mmcls/core/visualization/image.py", line 138, in wait_continue
    key_press = self.blocking_input(n=1, timeout=0)
  File "/home/user/.pyenv/versions/mmfewshot/lib/python3.8/site-packages/matplotlib/blocking_input.py", line 86, in __call__
    if self.figure.canvas.manager:
AttributeError: 'BlockingInput' object has no attribute 'figure'

Question about the training of MatchingNetwork

Hi, Great Job.

I have some questions about the training process of the matching network(classification)

• In this line,

  mmfewshot/configs/classification/matching_net/mini_imagenet/matching-net_conv4_1xb105_mini-imagenet_5way-1shot.py

  Line 28 in 31583cc

  num_shots=5,

  
  You use num_shots=5 for training 5-way-1-shot, is this a bug?
• The batch size shown in the result table is 64, I would like to know whether this number is the training batch size or test batch size?
• How many gaps between the meta-val and meta-test split in your experiment?
  • In the log of matching_net 5-way-1-shot, the max accuracy is about 51%, while the test result is 53%, does it means there exists ~2 points between two sets?

Thanks,
Best

Can't open the install.md and getting_started.md.

Can't open the install.md and getting_started.md.

CPU overload

When running program CPU is heavily overload which is not gonna happen on MMDETECTION(in my experience)?

ImportError: cannot import name 'DistributedSampler' from 'mmcls.datasets.builder'

使用tools/detection/test.py对TFA模型评估时发生如上报错，检查源码发现在mmfewshot/init.py中有from .classification import *一行，在后续调用mmcls的mmcls.datasets.builder文件中没有实现DistributedSampler。只有把from .classification import * 注释掉才能正常对detection模型进行测试

Model deployment

Hi, do you plan to support model deployment like in mmcls any time soon?

Can't install mmfewshot correctly with local libraies mmcls and mmdet.

I install MMClassification and MMDetection in the local library with git according to the get_start in MMDET and MMCLS.

ImportError: cannot import name 'DistributedSampler' from 'mmcls.datasets.builder' (/my/insttal/path/mmclassification/mmcls/datasets/builder.py)

loss nan when training fine-tuning based methods

Thanks for your code.
When i try the fine-tuning based methods such as TFA and FSCE, the loss become nan at about 250 iters. I tried by decreasing the lr, but it doesn't work. Could you give some suggestions?

I only have one GPU, Do I need to change lr = 0.001 to 0.000125？

I only have one GPU, Do I need to change lr = 0.001 to 0.000125？
Do samples_ per_ gpu = 2 need to be change？

coco dataset？

我的coco数据目录是这样的：
data
--coco
----annotations
----train2014
----val2014
--few_shot_ann
----coco
------benchmark_10shot
-------- ...
当我运行fsce下的coco预训练config时，会报错：no such file or directory: 'data/few_shot_ann/coco/annotaions/train.json'
请问这个train.json是哪里来的，预训练的标签不是应该调用coco文件夹下的annotations吗？
另外我在data preparation找到一个trainvalno5k.json和5k.json，请问是这两个json文件吗？
期待您的回答！

FileNotFoundError: FewShotCocoDefaultDataset: [Errno 2] No such file or directory: 'data/few_shot_ann/coco/benchmark_10shot/full_box_10shot_OD_trainval.json'

I use my own coco-fromat datasets, and then it shows the error. I need to spilt the all datasets? And is there some tools to split?

dataset

when train with my own datasets,a problem occur as follow:
Traceback (most recent call last):
File "/home/dio/VSST/anaconda3/envs/openmmlab/lib/python3.7/site-packages/mmcv/utils/registry.py", line 52, in build_from_cfg
return obj_cls(**args)
File "/home/dio/VSST/zwm/mmfewshot/mmfewshot/detection/datasets/voc.py", line 132, in init
**kwargs)
File "/home/dio/VSST/zwm/mmfewshot/mmfewshot/detection/datasets/base.py", line 140, in init
self.data_infos = self.ann_cfg_parser(ann_cfg)
File "/home/dio/VSST/zwm/mmfewshot/mmfewshot/detection/datasets/base.py", line 279, in ann_cfg_parser
return self.load_annotations(ann_cfg)
File "/home/dio/VSST/zwm/mmfewshot/mmfewshot/detection/datasets/voc.py", line 223, in load_annotations
ann_cfg_['ann_file'], ann_classes)
File "/home/dio/VSST/zwm/mmfewshot/mmfewshot/detection/datasets/voc.py", line 279, in load_annotations_xml
tree = ET.parse(xml_path)
File "/home/dio/VSST/anaconda3/envs/openmmlab/lib/python3.7/xml/etree/ElementTree.py", line 1197, in parse
tree.parse(source, parser)
File "/home/dio/VSST/anaconda3/envs/openmmlab/lib/python3.7/xml/etree/ElementTree.py", line 598, in parse
self._root = parser._parse_whole(source)
xml.etree.ElementTree.ParseError: no element found: line 1, column 0

During handling of the above exception, another exception occurred:

do you know what wrong with it?thank you!

meta_test_head is None on demo

The error occurs when running demo_metric_classifier_1shot_inference with a custom trained NegMargin model. The meta_test_head is None. Testing the model with dist_test works as expected though. I am not sure why it didn't save the meta test head. A comment here says that it is only built and run on testing. I am not sure what that means though.

The model config is the same as the standard in other config files:

model = dict(
    type='NegMargin',
    backbone=dict(type='Conv4'),
    head=dict(
        type='NegMarginHead',
        num_classes=6,
        in_channels=1600,
        metric_type='cosine',
        margin=-0.01,
        temperature=10.0),
    meta_test_head=dict(
        type='NegMarginHead',
        num_classes=6,
        in_channels=1600,
        metric_type='cosine',
        margin=0.0,
        temperature=5.0))

Otherwise, the config file itself is similar to other neg_margin config files for the cube dataset.

How to train custom dataset with same class in train-val-test split?

Hi!
I want to use few shot learning method to train model on my own dataset, but the dataset only has two classes, which means class for train-val-test from same class. How to train such data set with mmfewshot?

About custom datasets

I want to use mmfewshot to perform fewshot detection in a custom data set (coco format), but the guidance documents related to the use of custom data sets are too vague. I really can't figure out how to divide the data set, put it to several folders, and I don't know which file to change the settings in. I hope that the guidance document in this regard can be written in more detail, thank you.

Use Lvis dataset

Hello, your project is so great, but some papers use Lvis dataset for experiments, I wonder if you can add the usage of Lvis dataset into the project, thank you very much!

RuntimeError: The expanded size of the tensor (21) must match the existing size (54) at non-singleton dimension 0. Target sizes: [21, 1024]. Tensor sizes: [54, 1024]

Traceback (most recent call last): File "/home/lbc/miniconda3/envs/mmfewshot/lib/python3.7/runpy.py", line 193, in _run_module_as_main "__main__", mod_spec) File "/home/lbc/miniconda3/envs/mmfewshot/lib/python3.7/runpy.py", line 85, in _run_code exec(code, run_globals) File "/home/lbc/mmfewshot-main/tools/detection/misc/initialize_bbox_head.py", line 289, in <module> main() File "/home/lbc/mmfewshot-main/tools/detection/misc/initialize_bbox_head.py", line 278, in main args) File "/home/lbc/mmfewshot-main/tools/detection/misc/initialize_bbox_head.py", line 169, in random_init_checkpoint new_weight[:prev_cls] = pretrained_weight[:prev_cls] RuntimeError: The expanded size of the tensor (21) must match the existing size (54) at non-singleton dimension 0. Target sizes: [21, 1024]. Tensor sizes: [54, 1024]

The process of fsce on my own coco format datasets is:

1. Base Training : ckpt(step1)
2. step two: ues the best val pth of step 1 for train?
   python3.7 -m tools.detection.misc.initialize_bbox_head --src1 ./work_dirs/fsce_r101_fpn_coco_base-training/best_bbox_mAP_iter_105000.pth --method random_init --save-dir ./work_dirs/fsce_r101_fpn_coco-split1_base-training

save_best=True while encountering empty result

Describe the bug
Hi! I was deploying mmfewshot detection model on my own dataset. And "save_best=True" was added in evaluation dict.
Error happens for ckpt saving when empty result come out during validation stage

Error traceback
If applicable, paste the error trackback here.

here are some logs:

/home/user/anaconda3/envs/sc_mmlab/lib/python3.7/site-packages/mmcv/runner/hooks/evaluation.py:375: UserWarning: Since `eval_res` is an empty dict,
ior to save the best checkpoint will be skipped in this evaluation.                                                                                          
  'Since `eval_res` is an empty dict, the behavior to save '                                                                                                 
Traceback (most recent call last):                                                                                                                             
File "./Main/mmfewshot/tools/detection/train.py", line 215, in <module>                                                                                    
    main()                                                                                                                                                     
    File "./Main/mmfewshot/tools/detection/train.py", line 211, in main                                                                                        
    meta=meta)                                                                                                                                                 
    File "/home/user/sun_chen/Projects/FsDefect/Main/mmfewshot/mmfewshot/detection/apis/train.py", line 197, in train_detector                                 
    runner.run(data_loaders, cfg.workflow)                                                                                                                     
    File "/home/user/anaconda3/envs/sc_mmlab/lib/python3.7/site-packages/mmcv/runner/iter_based_runner.py", line 134, in run                                   
    iter_runner(iter_loaders[i], **kwargs)                                                                                                                     
    File "/home/user/anaconda3/envs/sc_mmlab/lib/python3.7/site-packages/mmcv/runner/iter_based_runner.py", line 67, in train                                  
    self.call_hook('after_train_iter')                                                                                                                         
    File "/home/user/anaconda3/envs/sc_mmlab/lib/python3.7/site-packages/mmcv/runner/base_runner.py", line 307, in call_hook                                   
    getattr(hook, fn_name)(self)                                                                                                                               
    File "/home/user/anaconda3/envs/sc_mmlab/lib/python3.7/site-packages/mmcv/runner/hooks/evaluation.py", line 262, in after_train_iter                       
    self._do_evaluate(runner)                                                                                                                                  
    File "/home/user/sun_chen/Projects/FsDefect/Main/mmfewshot/mmfewshot/detection/core/evaluation/eval_hooks.py", line 123, in _do_evaluate                   
    self._save_ckpt(runner, key_score)                                                                                                                       
    File "/home/user/anaconda3/envs/sc_mmlab/lib/python3.7/site-packages/mmcv/runner/hooks/evaluation.py", line 330, in _save_ckpt                                 
    if self.compare_func(key_score, best_score):                                                                                                             
    File "/home/user/anaconda3/envs/sc_mmlab/lib/python3.7/site-packages/mmcv/runner/hooks/evaluation.py", line 77, in <lambda>                                
    rule_map = {'greater': lambda x, y: x > y, 'less': lambda x, y: x < y}                                                                                   
TypeError: '>' not supported between instances of 'NoneType' and 'float'     

Bug fix

I noticed that codes in mmcv have already prevented this scenario from happening. BUT there features haven't been updated in mmfewshot and other mmlab repo. So small change following mmcv will save this.
turn

mmfewshot/mmfewshot/detection/core/evaluation/eval_hooks.py

Line 51 in 38afc01

if self.save_best:

and

mmfewshot/mmfewshot/detection/core/evaluation/eval_hooks.py

Line 124 in 38afc01

if self.save_best:

into

if self.save_best and key_score:

Visualisation of classification，t-sne or heapmap

Describe the feature
Dimensionality reduction of extracted features using t-sne. Visualisation of the extracted features of the feature extractor using class activation diagrams.

Motivation
For classification problems, visualization of feature extraction results is intuitive.

FSCE finetuning results with CPE loss on COCO

Hi, thank you for great codes! I find that there is no finetuning results of FSCE with CPE loss in the model zoo page (contrastive loss is None). Can you provide any training configurations with CPE loss? Thank you very much!

how to transfer attentionRPN to onnx format？

result index in show of test.py

https://github.com/open-mmlab/mmfewshot/blob/main/mmfewshot/detection/apis/test.py#L72

if is_module_wrapper(model):
    model.module.show_result(
        img_show,
        result[i],
        show=show,
        out_file=out_file,
        score_thr=show_score_thr)

should the index of result is j? result[j]?
I get an error: IndexError: list index out of range

NwayKshotDataset初始化num_used_support_shots=None报错

NwayKshot这个类里面，我把num_used_support_shots初始化None,但是调用prepare_support_shots方法的时候又对num_used_support_shot进行“<”运算；报错信息如下：

FewShotCocoDefaultDataset中coco_benchmark的ann_file路径无法自定义

在mmfewshot/detection/datasets/coco.py/FewShotCocoDefaultDataset 中的coco_benchmark固定了数据集路径为f'data/few_shot_ann/coco/benchmark_{shot}shot/full_box_{shot}shot_{class_name}_trainval.json'。但是我的few_shot_ann路径和上面不同，并且FewShotCocoDefaultDataset没有办法接受数据集路径的参数，希望可以增加此参数

problems about mini_imagenet dataset

The whole imagenet dataset must be downloaded from here if one want to use mini_imagenet.

But the mini_imagenet datasets popular in FSL community are often divided in advance, e.g. Optimization as a Model for Few-shot Learning

Hope this repo can support the mini_imagenet dataset from Optimization as a Model for Few-shot Learning

Are there any models in mmfewshot for segmentation task?

hello, thanks for your great work!
I learned a lot from this repo, and i wonder if there is any model here for segment task, which is a common and widely used task?

About the miniimagenet

Thank you for yourcontribution,after get the format of miniimage dataset,i train it,but a error occurs as follow:
2022-03-04 18:21:39,431 - mmfewshot - INFO - Set random seed to 0, deterministic: False
Traceback (most recent call last):
File "/home/dio/VSST/anaconda3/envs/openmmlab/lib/python3.7/site-packages/mmcv/utils/registry.py", line 52, in build_from_cfg
return obj_cls(**args)
File "/home/dio/VSST/zwm/mmfewshot/mmfewshot/classification/datasets/mini_imagenet.py", line 73, in init
super().init(*args, **kwargs)
File "/home/dio/VSST/zwm/mmfewshot/mmfewshot/classification/datasets/base.py", line 45, in init
self.data_infos = self.load_annotations()
File "/home/dio/VSST/zwm/mmfewshot/mmfewshot/classification/datasets/mini_imagenet.py", line 140, in load_annotations
int(filename.split('.')[0][-5:]) - 1]
IndexError: list index out of range
do you know what wrong with it?

about result reimplementation of meta-rcnn

When trying to reproduce results of meta-rcnn and TFA, under 1 shot setting of split1, I find that reproduced results of meta-rcnn is much higher, which is confusing.In paper of meta-rcnn(this 19.9 is the result i want to get):

In paper of TFA:

Result in paper shows that result of split1 under 1 shot setting is 19.9. But my results is much higher:
base training : mAP is 76.2
finetunning : all class is 47.40, novel class is 38.80, base class is 50.53
Which is much higher than results in paper. This is confusing. Besides, in the README.md of meta-rcnn, results are even higher:

under split1 1 shot setting, the results of TFA I get is 40.4 which is basically the same as the paper report.

Could you please kindly answer my questions?

IndexError: FewShotCocoDataset: list index out of range

train config:
python tools/detection/train.py configs/detection/fsce/coco/fsce_r101_fpn_coco_base-training.py

I want to train my own dataset with mmfewshot, so I modified the categories in BASE_CLASS in coco.py. The result is an error:

Traceback (most recent call last):
File "/home/anaconda3/envs/fewshot/lib/python3.7/site-packages/mmcv/utils/registry.py", line 52, in build_from_cfg
return obj_cls(**args)
File "/home/anaconda3/envs/fewshot/lib/python3.7/site-packages/mmfewshot/detection/datasets/coco.py", line 124, in init
**kwargs)
File "/home//anaconda3/envs/fewshot/lib/python3.7/site-packages/mmfewshot/detection/datasets/base.py", line 140, in init
self.data_infos = self.ann_cfg_parser(ann_cfg)
File "/home/anaconda3/envs/fewshot/lib/python3.7/site-packages/mmfewshot/detection/datasets/base.py", line 279, in ann_cfg_parser
return self.load_annotations(ann_cfg)
File "/home/anaconda3/envs/fewshot/lib/python3.7/site-packages/mmfewshot/detection/datasets/coco.py", line 196, in load_annotations
data_infos += self.load_annotations_coco(ann_cfg_['ann_file'])
File "/home/anaconda3/envs/fewshot/lib/python3.7/site-packages/mmfewshot/detection/datasets/coco.py", line 215, in load_annotations_coco
cat_id = self.coco.get_cat_ids(cat_names=[class_name])[0]
IndexError: list index out of range

How does it work

According to the document, the following errors occur during training. I don't know how to solve them. Has anyone encountered them.
TypeError: init() got an unexpected keyword argument 'persistent_workers'

confused about `samples_per_gpu` of meta_dataloader

mmfewshot/configs/detection/_base_/datasets/nway_kshot/base_voc.py

Line 106 in 486c8c2

samples_per_gpu=16,

Hi, thanks for your great work in fsod.
I want to know why the value of samples_per_gpu is not 15 instead of 16 for voc base training.
Hope you can help me.

Don't find the “frozen_parameters” parameter in the relevant source code

I found that the “frozen_parameters” parameter is used in many detection models, but I have not found where this parameter is used in the relevant source code. Which part of the source code should I see?

How to use the detectors in mmdetection?

I want to use some detectors in mmdet like detr, but many modules in detr don't be implemented in mmfewshot. Maybe I can use "custom import" in config to import every modules, however I want to consult that is there any other way to achieve it?

fatal: ambiguous argument 'HEAD': unknown revision or path not in the working tree. Use '--' to separate paths from revisions, like this: 'git <command> [<revision>...] -- [<file>...]'

我是用fsce的base训练，出现以下错误
fatal: ambiguous argument 'HEAD': unknown revision or path not in the working tree.
Use '--' to separate paths from revisions, like this:
'git [...] -- [...]'

貌似和git提交有关 但是我并不需要提交任何代码

IterBasedRunner chosen instead of EpochBasedRunner

Hi！ First of all, very appreciate of your excellent job and it really helps me a lot for my own research.

In the past few weeks, I was trying to re-implement some fsod work based on mmdet. Since MMFewShot was released, I spent a few days reading the source code, and noticed a little difference. In mmdet-style config file,, EpochBasedRunner is default runner type, but IterBasedRunner was chosen in this project.

I wonder if there is any help or rise on results through this change. If it really helps, do you have any suggestion on converting runner from EpochBasedRunner to IterBasedRunner, for example some empirical formula to calculate max_iter and evaluation interval between EpochBasedRunner and IterBasedRunner, since i would like to conduct fsod models on some domain-specific dataset.

Can't install mmfewshot correctly with openmim

$ pip install openmim
$ mim install mmfewshot
$ mim list

Package    Version    Source
---------  ---------  --------

$ python demo/demo_metric_classifier_1shot_inference.py \
    demo/demo_classification_images/query_images/Least_Auklet.jpg \
    configs/classification/proto_net/cub/proto-net_conv4_1xb105_cub_5way-1shot.py \
    https://download.openmmlab.com/mmfewshot/classification/proto_net/cub/proto-net_conv4_1xb105_cub_5way-1shot_20211120_101211-9ab530c3.pth \
    --support-images-dir demo/demo_classification_images/support_images

Traceback (most recent call last):
  File "demo/demo_metric_classifier_1shot_inference.py", line 14, in <module>
    from mmfewshot.classification.apis import (inference_classifier,
  File "/miniconda3/envs/mmfewshot/lib/python3.8/site-packages/mmfewshot/__init__.py", line 2, in <module>
    import mmcls  # noqa: F401, F403
ModuleNotFoundError: No module named 'mmcls'

visualization program

I wonder whether this code has not provided the code of the visualization program, using the TFA trained model for test? I have reproduced TFA and trained the model with this code. And I want to take an image to see the effect of the detection with visual interface.Could you help me？Thanks a lot！

使用pytorch-style的resnet训练

在mmfewshot中好像都是使用caffe-style的faster-rcnn，当我尝试将模型换成pytorch-style后训练fsce-voc-base train，第一个iter时loss就变成了nan，但是使用caffe-style则可以正常训练。pytorch-style的config如下：
base = ['../../base/models/faster_rcnn_r50_fpn.py',] #从mmdet的config中复制过来
model = dict(
backbone=dict(
depth=101,
norm_cfg=dict(type='BN', requires_grad=True),
init_cfg=dict(type='Pretrained', checkpoint='torchvision://resnet101')
),
roi_head=dict(bbox_head=dict(num_classes=15)))
请问问题可能出在哪里