Rebase upstream

.dockerignore

@@ -0,0 +1,2 @@
+runs
+weights

.env

@@ -0,0 +1 @@
+TAG=0.1.0

Dockerfile

@@ -0,0 +1,55 @@
+#FROM nvcr.io/nvidia/pytorch:20.06-py3
+#FROM continuumio/miniconda3
+#FROM gpuci/miniconda-cuda:11.0-devel-ubuntu20.04
+FROM nvidia/cuda:11.1-cudnn8-devel-ubuntu20.04
+
+#COPY environment.yml .
+#RUN conda update --force-reinstall conda
+#RUN conda env update --name base --file environment.yml --prune
+
+ENV MY_ROOT=/workspace \
+    PKG_PATH=/yolo_src \
+    NUMPROC=4 \
+    PYTHON_VER=3.8 \
+    PYTHONUNBUFFERED=1 \
+    PYTHONPATH=. \
+    DEBIAN_FRONTEND=noninteractive 
+
+WORKDIR $PKG_PATH
+
+RUN apt-get update && apt-get install -y apt-utils && apt-get -y upgrade && \
+    apt-get install -y git libsnappy-dev libopencv-dev libhdf5-serial-dev libboost-all-dev libatlas-base-dev \
+        libgflags-dev libgoogle-glog-dev liblmdb-dev curl unzip\
+        python${PYTHON_VER}-dev && \
+    curl https://bootstrap.pypa.io/get-pip.py -o get-pip.py && \
+    python${PYTHON_VER} get-pip.py && \
+    rm get-pip.py && \
+    # Clean UP
+    apt upgrade -y && \
+    apt clean && \
+    apt autoremove -y && \
+    rm -rf /var/lib/apt/lists/*  # cleanup to reduce image size
+
+RUN ln -s /usr/bin/python${PYTHON_VER} /usr/bin/python
+
+RUN apt install unzip
+RUN pip install torch==1.7.1+cu110 torchvision==0.8.2+cu110 -f https://download.pytorch.org/whl/torch_stable.html
+
+WORKDIR $MY_ROOT
+# We have to install mish-cuda from source due to an issue with one of the header files
+ADD https://github.com/thomasbrandon/mish-cuda/archive/master.zip $MY_ROOT/mish-cuda.zip
+RUN unzip mish-cuda.zip
+WORKDIR $MY_ROOT/mish-cuda-master
+RUN cp external/CUDAApplyUtils.cuh csrc/
+RUN python setup.py build install
+WORKDIR $PKG_PATH
+# ADD https://drive.google.com/file/d/1NQwz47cW0NUgy7L3_xOKaNEfLoQuq3EL/view?usp=sharing /weights/yolov4-csp.weights
+ADD requirements.txt $PKG_PATH/requirements.txt
+RUN pip install -r $PKG_PATH/requirements.txt
+ADD yolo $PKG_PATH/yolo
+ADD train.py $PKG_PATH/train.py
+ADD test.py $PKG_PATH/test.py
+ADD setup.py $PKG_PATH/setup.py
+ADD data $PKG_PATH/data
+RUN pip install .
+

README.md

@@ -1 +1,110 @@
-# scaledYOLOv4
+# YOLOv4-CSP
+
+This is the implementation of "[Scaled-YOLOv4: Scaling Cross Stage Partial Network](https://arxiv.org/abs/2011.08036)" using PyTorch framwork.
+
+* **2021.05.21** Due to unknown issue some people can not reproduce the performance in paper and I can not reproduce the [issue#89](https://github.com/WongKinYiu/ScaledYOLOv4/issues/89), I update the codebase. But it will makes the reproduce performance becomes better than paper (47.8 AP -> 48.7 AP).
+
+* **2020.11.16** Now supported by [Darknet](https://github.com/AlexeyAB/darknet). [`yolov4-csp.cfg`](https://github.com/AlexeyAB/darknet/blob/master/cfg/yolov4-csp.cfg) [`yolov4-csp.weights`](https://drive.google.com/file/d/1TdKvDQb2QpP4EhOIyks8kgT8dgI1iOWT/view?usp=sharing)
+
+## Installation
+
+```
+# create the docker container, you can change the share memory size if you have more.
+nvidia-docker run --name yolov4_csp -it -v your_coco_path/:/coco/ -v your_code_path/:/yolo --shm-size=64g nvcr.io/nvidia/pytorch:20.11-py3
+
+# apt install required packages
+apt update
+apt install -y zip htop screen libgl1-mesa-glx
+
+# pip install required packages
+pip install seaborn thop
+
+# install mish-cuda if you want to use mish activation
+# https://github.com/thomasbrandon/mish-cuda
+# https://github.com/JunnYu/mish-cuda
+cd /
+git clone https://github.com/JunnYu/mish-cuda
+cd mish-cuda
+python setup.py build install
+
+# go to code folder
+cd /yolo
+```
+
+## Testing
+
+[`yolov4-csp.weights`](https://drive.google.com/file/d/1TdKvDQb2QpP4EhOIyks8kgT8dgI1iOWT/view?usp=sharing)
+
+<details><summary> <b>old weights</b> </summary>
+
+[`yolov4-csp.weights`](https://drive.google.com/file/d/1NQwz47cW0NUgy7L3_xOKaNEfLoQuq3EL/view?usp=sharing)
+
+</details>
+
+```
+# download yolov4-csp.weights and put it in /yolo/weights/ folder.
+python test.py --img 640 --conf 0.001 --iou 0.65 --batch 8 --device 0 --data coco.yaml --cfg models/yolov4-csp.cfg --weights weights/yolov4-csp.weights
+```
+
+You will get the results:
+```
+ Average Precision  (AP) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.48656
+ Average Precision  (AP) @[ IoU=0.50      | area=   all | maxDets=100 ] = 0.67002
+ Average Precision  (AP) @[ IoU=0.75      | area=   all | maxDets=100 ] = 0.52739
+ Average Precision  (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.33082
+ Average Precision  (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.54036
+ Average Precision  (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.62107
+ Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=  1 ] = 0.37197
+ Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets= 10 ] = 0.61211
+ Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.66544
+ Average Recall     (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.49676
+ Average Recall     (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.72018
+ Average Recall     (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.80528
+```
+<details><summary> <b>old results</b> </summary>
+
+```
+ Average Precision  (AP) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.47827
+ Average Precision  (AP) @[ IoU=0.50      | area=   all | maxDets=100 ] = 0.66448
+ Average Precision  (AP) @[ IoU=0.75      | area=   all | maxDets=100 ] = 0.51928
+ Average Precision  (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.30647
+ Average Precision  (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.53106
+ Average Precision  (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.61056
+ Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=  1 ] = 0.36823
+ Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets= 10 ] = 0.60434
+ Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.65795
+ Average Recall     (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.48486
+ Average Recall     (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.70892
+ Average Recall     (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.79914
+```
+
+</details>
+
+## Training
+
+```
+# you can change batch size to fit your GPU RAM.
+python train.py --device 0 --batch-size 16 --data coco.yaml --cfg yolov4-csp.cfg --weights '' --name yolov4-csp
+```
+
+For resume training:
+```
+# assume the checkpoint is stored in runs/train/yolov4-csp/weights/.
+python train.py --device 0 --batch-size 16 --data coco.yaml --cfg yolov4-csp.cfg --weights 'runs/train/yolov4-csp/weights/last.pt' --name yolov4-csp --resume
+```
+
+If you want to use multiple GPUs for training
+```
+python -m torch.distributed.launch --nproc_per_node 4 train.py --device 0,1,2,3 --batch-size 64 --data coco.yaml --cfg yolov4-csp.cfg --weights '' --name yolov4-csp --sync-bn
+```
+
+## Citation
+
+```
+@article{wang2020scaled,
+  title={{Scaled-YOLOv4}: Scaling Cross Stage Partial Network},
+  author={Wang, Chien-Yao and Bochkovskiy, Alexey and Liao, Hong-Yuan Mark},
+  journal={arXiv preprint arXiv:2011.08036},
+  year={2020}
+}
+```

data/coco.names

@@ -0,0 +1,80 @@
+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

data/coco.yaml

@@ -0,0 +1,19 @@
+# train and val datasets (image directory or *.txt file with image paths)
+train: /mnt/coco/5k.txt  # 118k images
+#val: /mnt/coco/5k.txt  # 5k images
+val: data/coco_5k_500_subset.txt
+test: data/coco_5k_500_subset.txt  # 20k images for submission to https://competitions.codalab.org/competitions/20794
+
+# number of classes
+nc: 80
+
+# class names
+names: ['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']

data/hyp.scratch.yaml

@@ -0,0 +1,28 @@
+lr0: 0.01  # initial learning rate (SGD=1E-2, Adam=1E-3)
+lrf: 0.2  # final OneCycleLR learning rate (lr0 * lrf)
+momentum: 0.937  # SGD momentum/Adam beta1
+weight_decay: 0.0005  # optimizer weight decay 5e-4
+warmup_epochs: 3.0  # warmup epochs (fractions ok)
+warmup_momentum: 0.8  # warmup initial momentum
+warmup_bias_lr: 0.1  # warmup initial bias lr
+box: 0.05  # box loss gain
+cls: 0.3  # cls loss gain
+cls_pw: 1.0  # cls BCELoss positive_weight
+obj: 0.7  # obj loss gain (scale with pixels)
+obj_pw: 1.0  # obj BCELoss positive_weight
+iou_t: 0.20  # IoU training threshold
+anchor_t: 4.0  # anchor-multiple threshold
+# anchors: 3  # anchors per output layer (0 to ignore)
+fl_gamma: 0.0  # focal loss gamma (efficientDet default gamma=1.5)
+hsv_h: 0.015  # image HSV-Hue augmentation (fraction)
+hsv_s: 0.7  # image HSV-Saturation augmentation (fraction)
+hsv_v: 0.4  # image HSV-Value augmentation (fraction)
+degrees: 0.0  # image rotation (+/- deg)
+translate: 0.1  # image translation (+/- fraction)
+scale: 0.9  # image scale (+/- gain)
+shear: 0.0  # image shear (+/- deg)
+perspective: 0.0  # image perspective (+/- fraction), range 0-0.001
+flipud: 0.0  # image flip up-down (probability)
+fliplr: 0.5  # image flip left-right (probability)
+mosaic: 1.0  # image mosaic (probability)
+mixup: 0.0  # image mixup (probability)

data/speedco.yaml

@@ -0,0 +1,10 @@
+# train and val datasets (image directory or *.txt file with image paths)
+train: data/speedco_dataset_train_images.txt
+val: data/speedco_dataset_val_images.txt
+test: data/speedco_dataset_val_images.txt
+
+# number of classes
+nc: 1
+
+# class names
+names: ['truck']