kkksqj/deeplearning

A deep learning code base, mainly for paper replication, in the areas of image recognition, object detection, image segmentation, self-supervision, etc. Each project can be run independently, and there are corresponding articles to explain.

DeepLearning

介绍

个人学习项目，用于复现论文代码，深入理解算法原理。

注意：每个项目都可独立运行。若要运行某个项目，你需要将该项目作为根目录，以便找到对应模块。

每个项目都有对应论文解读，解读详情搜 [知乎] - 琪小钧

• paper read: 知乎 论文解读（有些博主对于某些论文已经作了很深刻的理解，因此有些算法直接引用了他们的知乎文章。如有处理不当的地方，请联系我。）
• code:对应项目代码

classification

• mnist paper read / code
• vggNet paper read / code
• resnet paper read / code
• resNeXt paper read / code
• resNeSt paper read / code
• coatNet paper read / code
• convNext paper read / code
• seNet paper read / code
• skNet paper read / code
• TransFG paper read / code
• RepVGG paper read / code
• efficientNet paper read / code
• shuffleNet paper read / code
• vision_transformer paper read / code
• swin-transformer paper read / code
• googleNet paper read / code
• inception
• denseNet
• CBAM
• mobileNet
• Xception
• SqueezeNet

detection

• FPN(实现resnet50 + fpn) paper read / code
• Faster-rcnn paper read / code
• yolov7 paper read / code
• RetinaNet (包含focal_loss) code
• YOLOV5 V5.0 (实现注释，更新pt->onnx代码) code
• yolox (修改了voc数据读取方式) code
• FCOS code
• yoloF
• yoloR
• detr
• ssd
• Mask-rcnn
• Cascade-rcnn
• SPPNet
• CenterNet
• RepPoints
• OTA
• ATSS

segmentation

• U-Net paper read / code
• HR-Net-Seg paper read / code
• DeepLabv3 paper read / code
• DeepLabv3Plus paper read / code
• few_shot segmentation code
• FCN code
• Mask-rcnn
• Cascade-rcnn
• UNet++
• PSPNet
• Segmenter

metric_learning

• BDB (用于图像检索) code
• Happy-Whale (鲸鱼竞赛检索baseline) code

self-supervised

• MAE (实现VIT+MAE) paper read / code
• SupCon (实现自对比学习+t-SNE可视化+swa) paper read / code
• MoCo
• SwAV

deep_stereo

• Real_time_self_adaptive_deep_stereo (实时双目里立体匹配，细节待完善) code

other

• label_convert (三种不同标注文件之间的转换以及box可视化) paper read / code
• normalization (BN、LN、IN、GN、SN图解) paper read / code
• DDP (模型分布式计算) paper read / code
• tensorboard test (可视化网络，图片，训练过程以及卷积核) / code
• load weights test (权重部分加载) / code
• visual weights map test (特征图、卷积核可视化分析) / code
• class_Activation_Map_Visual (可视化CNN的类激活图) / code
• deploy (pytorch模型转onnx，支持自定义算子 示例) / code

deeplearning's Issues

Loss does not decrease while training

Hi，thanks a lot for publishing the code. I had a question: I use your code to train CUB_200_2011datasets, but loss does not decrease while training

GPU指定

使用vggnet，有多块GPU怎么指定具体某一块？

大佬coco转yolo之后，yolo的.yaml文件的类别怎么配置啊？

转了之后不知道怎么配置datasets下自定义的.yaml文件，
names:
0: Drosicha_contrahens_female
1: Drosicha_contrahens_male
2: Chalcophora_japonica
3: Anoplophora_chinensis
4: Psacothea_hilaris(Pascoe)
5: Apriona_germari(Hope)
6: Monochamus_alternatus
7: Plagiodera_versicolora(Laicharting)
8: Latoia_consocia_Walker
9: Hyphantria_cunea
10: Cnidocampa_flavescens(Walker)
11: Cnidocampa_flavescens(Walker_pupa)
......
不知道怎么能抽取出来，然后运行会报错
Traceback (most recent call last):
File "train.py", line 8, in
model.train(data="./ultralytics/cfg/datasets/pest.yaml", epochs=3) # train the model
File "/root/autodl-tmp/ultralytics-main/ultralytics/engine/model.py", line 336, in train
self.trainer = (trainer or self._smart_load('trainer'))(overrides=args, _callbacks=self.callbacks)
File "/root/autodl-tmp/ultralytics-main/ultralytics/engine/trainer.py", line 123, in init
raise RuntimeError(emojis(f"Dataset '{clean_url(self.args.data)}' error ❌ {e}")) from e
RuntimeError: Dataset 'ultralytics/cfg/datasets/pest.yaml' error ❌ object of type 'NoneType' has no len()

TransFG on stcar dataset

你好，请问你们是否尝试过在斯坦福汽车数据集上测试代码，我用源码跑下来效果一直很差，比论文中的效果低3%左右

能加个微信吗大佬

我的微信号NatBiotechnol

关于用IP102数据集训练遇到的问题

您好作者，我在用TransFG跑IP102数据集的时候，在训练阶段loss一直保持在4点多不变，Valid Accuracy只有百分之0.4，经过多轮训练一直不变化，请问您有遇到过这类的问题么？

大佬，binary mask 是什么格式的数据集，要怎么生成这样的掩码？谢谢

大佬你好，请问vgg数据集可以用CPU训练吗？谢谢