luogen1996 / oneteacher Goto Github PK

License: GNU General Public License v3.0

Dockerfile 0.44% Shell 1.16% Python 98.40%

oneteacher's Introduction

OneTeacher

Official implementation of "Towards End-to-end Semi-supervised Learning for One-stage Object Detection". OneTeacher is a semi-supervised framework for YOLOV5, which is equiped with two novel designs, namely Multi-view Pseudo-label Refinement and Decoupled Semi-supervised Optimization.

Updates

(2023/1/29) Release our project.

Data Preparation

Prepare the datasets of COCO and VOC, please refer to YOLOv5

After that, the file structure should look like:

   |-- OneTeacher
   |-- datasets
       |-- coco
             |-- images
                  |-- train2017
                  |-- val2017
                  |-- test2017
             |-- labels
             |-- annotations
                  |-- instances_train2017.json
                  |-- instances_val2017.json

Generate semi-supervised data split.

python active_sampling/generate_random_supervised_seed_yolo.py  --dataset_name 'coco_2017_train'  --random_seeds 0,1,2,3,4,5,6,7,8,9  --random_file  ./data_processing/COCO_supervision.txt   --random_percent 10.0  --output_file ./dataseed/COCO_supervision_10.json

Training and Evaluation

Semi-supervised Learning on COCO 10%

bash script/coco_semi_script.sh 32 2 0,1

Semi-supervised Learning on VOC 25%

bash script/voc_semi_script.sh 32 2 0,1

Fully-supervised Learning on COCO 10%

bash script/coco_fully_script.sh 32 2 0,1

Fully-supervised Learning on VOC 25%

bash script/voc_fully_script.sh 32 2 0,1

oneteacher's People

Contributors

Stargazers

Watchers

Forkers

jie311 hiwyl xia-tao cv-det lonelyzyp jiajiali04 light-cat haoyue-code 3321zsc

oneteacher's Issues

About pre-trained weight

Can you provide a pre-trained weight?@ @luogen1996

do you know how to solve this issue "NameError: name '_SPLITS_COCO5_FORMAT' is not defined" ?

when i execution this command ,

"python data_processing/generate_random_supervised_seed_yolo.py --dataset_name 'coco_2017_train' --random_seeds 0,1,2,3,4,5,6,7,8,9 --random_file ./data_processing/COCO_supervision.txt --random_percent 10.0 --output_file ./dataseed/COCO_supervision_10.json"

this error was occured .

"Traceback (most recent call last):
File "data_processing/generate_random_supervised_seed_yolo.py", line 7, in
from builtin import *
File "/home/ssod/OneTeacher/data_processing/builtin.py", line 175, in
register_coco5(_root)
File "/home/ssod/OneTeacher/data_processing/builtin.py", line 115, in register_coco5
for _, splits_per_dataset in _SPLITS_COCO5_FORMAT.items():
NameError: name '_SPLITS_COCO5_FORMAT' is not defined"

please help me.

how to use OneTeacher for custom data?

About train_semi.py

When i excute the script, it met the problem(it appeared epoch 12):
Traceback (most recent call last): File "train_20.py", line 920, in <module> main(opt) File "train_20.py", line 818, in main train(opt.hyp, opt, device, callbacks) File "train_20.py", line 511, in train out, train_out,pseudo_class_one_hot =model_teacher(unlabel_imgs_weak_aug,augment=True) File "/home/yy/miniconda3/envs/general/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1110, in _call_impl return forward_call(*input, **kwargs) File "/home/yy/log/4.semi-supervised/OneTeacher-main/models/yolo.py", line 523, in forward return self.forward_augment(x) # augmented inference, None File "/home/yy/log/4.semi-supervised/OneTeacher-main/models/yolo.py", line 533, in forward_augment xi = scale_img(x.flip(fi) if fi else x, si, gs=int(self.stride.max())) File "/home/yy/log/4.semi-supervised/OneTeacher-main/utils/torch_utils.py", line 300, in scale_img s = (int(h * ratio), int(w * ratio)) # new size TypeError: unsupported operand type(s) for *: 'int' and 'NoneType'@ @luogen1996

关于论文中的一个疑问

论文中这张图表示了在使用ImageNet pre-training后的one-teacher和其它方法结果的比较，但是论文中前面写道one-teacher使用的是yolov5-s作为base model。yolo的backbone结构和resnent差距挺大的，如果我想复现这部分实验结果，请问我该如何使用ImageNet的预训练权重来初始化yolov5-s？

About non_max_suppression_pseudo_decouple_multi_view

if isinstance(out,list):
 
pseudo_boxes_reg, pseudo_boxes_cls = non_max_suppression_pseudo_decouple_multi_view(out, hyp[ 'bbox_threshold'], hyp['cls_threshold'], multi_label=True ,agnostic=single_cls, )

else:
 
pseudo_boxes_reg,pseudo_boxes_cls = non_max_suppression_pseudo_decouple(out, hyp['bbox_threshold'],hyp['cls_threshold'], multi_label=True, agnostic=single_cls )

OneTeacher/train_semi.py

Line 527 in 9a0e798

if isinstance(out,list):

When I run code, it always run at non_max_suppression_pseudo_decouple. How type of model, how type config can output a list and run non_max_suppression_pseudo_decouple_multi_view?

how could i implement one teacher with YOLOv7 ?

@luogen1996 which are the main steps to be able to integrate yolov7 in place of yolov5 ?

按10% 有标注，90%无标注训练coco 时，使用non_max_suppression_pseudo_decouple_multi_view 产生伪标签时，这个函数中的labels真实标签被赋值，这是否已经不是半监督了？

想知道论文中的实验结论是否是在labels 为None 的情况下得出的。

train_semi error about student model !!!!!

after resolving all problems about training my custom dataset with One teacher when I launch the training, I got this error:

 Epoch   gpu_mem       box       obj       cls       mls  semi_obj  semi_cls  semi_mls    labelssemi_labels
 0/299     2.68G    0.1051   0.09514   0.01167   0.09257         0         0         0       104         0:  22%|███████▊                            | 3000/13860 [02:58<10:45, 16.81it/s]

Traceback (most recent call last):
File "/app/train_semi.py", line 931, in
main(opt)
File "/app/train_semi.py", line 829, in main
train(opt.hyp, opt, device, callbacks)
File "/app/train_semi.py", line 510, in train
_update_teacher_model(model if ema is None else ema.ema, model_teacher, word_size=WORLD_SIZE, keep_rate=0.)
File "/opt/conda/lib/python3.10/site-packages/torch/utils/_contextlib.py", line 115, in decorate_context
return func(*args, **kwargs)
File "/app/utils/torch_utils.py", line 48, in _update_teacher_model
raise Exception("{} is not found in student model".format(key))
Exception: model.0.conv.conv.weight is not found in student model

How could I resolve it ?

ModuleNotFoundError: No module named 'detectron2'

why？

About data_processing/generate_random_supervised_seed_yolo.py

发生异常: NameError name '_SPLITS_COCO5_FORMAT' is not defined File "/home/cv/yj/5.semi-object/OneTeacher-main/data_processing/builtin.py", line 103, in register_coco5 for _, splits_per_dataset in _SPLITS_COCO5_FORMAT.items(): File "/home/cv/yj/5.semi-object/OneTeacher-main/data_processing/builtin.py", line 160, in <module> register_coco5(_root) File "/home/cv/yj/5.semi-object/OneTeacher-main/data_processing/generate_random_supervised_seed_yolo.py", line 7, in <module> from builtin import * NameError: name '_SPLITS_COCO5_FORMAT' is not defined
generate_random_supervised_seed_yolo.py: error: unrecognized arguments: --output_file, how to generate the COCO_supervision_10.json, what is the form of COCO_supervision_10.json?
where is the active_sampling directory, data_processing directory denotes active_sampling directory?
If i want to use own datasets, how can i use generate_random_supervised_seed_yolo.py? it exists some problems Attribute 'thing_classes' in the metadata of 'coco_2017_train' cannot be set to a different value! ['person', 'bicycle', 'car', 'motorcycle', 'airplane', 'bus', 'train', 'truck', 'boat', 'traffic light', 'fire hydrant', 'stop sign', 'parking meter', 'bench', 'bird', 'cat', 'dog', 'horse', 'sheep', 'cow', 'elephant', 'bear', 'zebra', 'giraffe', 'backpack', 'umbrella', 'handbag', 'tie', 'suitcase', 'frisbee', 'skis', 'snowboard', 'sports ball', 'kite', 'baseball bat', 'baseball glove', 'skateboard', 'surfboard', 'tennis racket', 'bottle', 'wine glass', 'cup', 'fork', 'knife', 'spoon', 'bowl', 'banana', 'apple', 'sandwich', 'orange', 'broccoli', 'carrot', 'hot dog', 'pizza', 'donut', 'cake', 'chair', 'couch', 'potted plant', 'bed', 'dining table', 'toilet', 'tv', 'laptop', 'mouse', 'remote', 'keyboard', 'cell phone', 'microwave', 'oven', 'toaster', 'sink', 'refrigerator', 'book', 'clock', 'vase', 'scissors', 'teddy bear', 'hair drier', 'toothbrush'] != ['graminearum', 'pseudograminearum', 'moniliforme', 'glycines', 'gloeosporioides', 'clavispora', 'anthracnose'] File "/home/cv/yj/5.semi-object/detectron2/detectron2/data/catalog.py", line 148, in __setattr__ assert oldval == val, ( File "/home/cv/yj/5.semi-object/detectron2/detectron2/data/datasets/coco.py", line 80, in load_coco_json meta.thing_classes = thing_classes File "/home/cv/yj/5.semi-object/detectron2/detectron2/data/datasets/coco.py", line 500, in <lambda> DatasetCatalog.register(name, lambda: load_coco_json(json_file, image_root, name)) File "/home/cv/yj/5.semi-object/detectron2/detectron2/data/catalog.py", line 58, in get return f() File "/home/cv/yj/5.semi-object/detectron2/detectron2/data/build.py", line 241, in <listcomp> dataset_dicts = [DatasetCatalog.get(dataset_name) for dataset_name in names] File "/home/cv/yj/5.semi-object/detectron2/detectron2/data/build.py", line 241, in get_detection_dataset_dicts dataset_dicts = [DatasetCatalog.get(dataset_name) for dataset_name in names] File "/home/cv/yj/5.semi-object/OneTeacher-main/data_processing/generate_random_supervised_seed_yolo.py", line 11, in generate dataset_dicts = get_detection_dataset_dicts( File "/home/cv/yj/5.semi-object/OneTeacher-main/data_processing/generate_random_supervised_seed_yolo.py", line 55, in <module> generate( AssertionError: Attribute 'thing_classes' in the metadata of 'coco_2017_train' cannot be set to a different value! ['person', 'bicycle', 'car', 'motorcycle', 'airplane', 'bus', 'train', 'truck', 'boat', 'traffic light', 'fire hydrant', 'stop sign', 'parking meter', 'bench', 'bird', 'cat', 'dog', 'horse', 'sheep', 'cow', 'elephant', 'bear', 'zebra', 'giraffe', 'backpack', 'umbrella', 'handbag', 'tie', 'suitcase', 'frisbee', 'skis', 'snowboard', 'sports ball', 'kite', 'baseball bat', 'baseball glove', 'skateboard', 'surfboard', 'tennis racket', 'bottle', 'wine glass', 'cup', 'fork', 'knife', 'spoon', 'bowl', 'banana', 'apple', 'sandwich', 'orange', 'broccoli', 'carrot', 'hot dog', 'pizza', 'donut', 'cake', 'chair', 'couch', 'potted plant', 'bed', 'dining table', 'toilet', 'tv', 'laptop', 'mouse', 'remote', 'keyboard', 'cell phone', 'microwave', 'oven', 'toaster', 'sink', 'refrigerator', 'book', 'clock', 'vase', 'scissors', 'teddy bear', 'hair drier', 'toothbrush'] != ['grami', 'pseudogrami', 'monili', 'glyci', 'gloeos', 'clavis', 'anth']

没有COCO_supervision_10.json文件生成

甚至generate_random_supervised_seed_yolo.py里面都没有output_file这个参数。代码是不是缺了啊 @luogen1996