Table of Contents

• Report
• How to reproduce results
• How to evaluate a model different from yolox
• How to setup Euler
• Euler commands

Report

TODO

How to reproduce results

Installation

Step 1: Install YOLOX-Bees.

git clone https://github.com/AlessandroRuzzi/YOLOX-Bees
cd YOLOX-Bees
pip3 install -U pip && pip3 install -r requirements.txt
pip3 install -v -e .  # or  python3 setup.py develop

Step 2: Install pycocotools.

pip3 install cython; pip3 install 'git+https://github.com/cocodataset/cocoapi.git#subdirectory=PythonAPI'

Training

Step 1: Download a yolox pre-trained checkpoint from the table below and put it in the folder YOLOX-Bees/checkpoints.

Standard Models.

Model	size	mAPval 0.5:0.95	mAPtest 0.5:0.95	Speed V100 (ms)	Params (M)	FLOPs (G)	weights
YOLOX-s	640	40.5	40.5	9.8	9.0	26.8	github
YOLOX-m	640	46.9	47.2	12.3	25.3	73.8	github
YOLOX-l	640	49.7	50.1	14.5	54.2	155.6	github
YOLOX-x	640	51.1	51.5	17.3	99.1	281.9	github

Step 2: Based on the checkpoint you downloaded you will choose a different experiment file. They are located in /YOLOX-Bees/exps/default/ and you can choose between yolox_s, yolox_m, yolox_l and yolox_x.

Step 3: Download from Azure the zip file /beelivingsensor/dslab2021/dslab2020_bee_detection_data_blurred/reproduce_results_dataset.zip, unzip it and put all the folders inside the folder YOLOX_Bees/datasets/.

Step 4: Run the following command to train yolox using a single GPU (it can only be trained with GPUs)

python tools/train.py -f exps/default/YOUR_EXP_FILE.py -d 1 -b 4 --fp16 -o -c checkpoints/YOUR_CHECKPOINT.pth

If you are using Euler cluster you can run:

bsub -W 24:00 -o log_test -R "rusage[mem=32000, ngpus_excl_p=1]" -R "select[gpu_model0==GeForceRTX2080Ti]" python tools/train.py -f exps/default/YOUR_EXP_FILE.py -d 1 -b 4 --fp16 -o -c checkpoints/YOUR_CHECKPOINT.pth

• -d: number of gpu devices
• -b: total batch size, the recommended number for -b is num-gpu * 8
• --fp16: mixed precision training

Step 5: Once the train ends (around 80/100 epochs should be enough) you will find in the folder /YOLOX-Bees/YOLOX_outputs/YOUR_EXP_NAME/ the best checkpoint (evaluated on the validation set) and the last epoch checkpoint.

Evaluation

Step 1: Download a yolox checkpoint from Azure /beelivingsensor/dslab2021/dslab2020_bee_detection_data_blurred/best_ckpt_yolox_l.pth or use one checkpoint that you produced and put it in the folder YOLOX-Bees/checkpoints/.

Step 2: Then if you haven't already done it, download from Azure the zip file /beelivingsensor/dslab2021/dslab2020_bee_detection_data_blurred/reproduce_results_dataset.zip, unzip it and put all the folders inside the folder YOLOX_Bees/datasets/.

Step 3: Open the file YOLOX-Bees/exps/default/yolox_bees_eval.py and modify self.depth and self.width based on the checkpoint you have downloaded ( yolox_x -> [1.33, 1.25] , yolox_l -> [1.0, 1.0] , yolox_m -> [0.67, 0.75] , yolo_s -> [0.33, 0.50]] )

Step 4: Run the following command to obtain predictions for all the datasets

python evaluation.py image -f exps/default/yolox_bees_eval.py -c checkpoints/YOUR_CHECKPOINT.pth --tsize 832 --save_result --conf 0.05 --nms 0.8

Step 5: At the end you will find a file called mAP_results.txt together with an output file for each dataset in the folder YOLOX-Bees/map/output/, while you will find images with bounding boxes predicted by the model in the folder YOLOX-Bees/YOLOX_outputs/yolox_bees_eval/.

How to evaluate a model different from yolox

Create the detection-results files

Step 1: Use your model to create a separate detection-results text file for each image for each dataset.

Step 2: Use matching names for the files (e.g. image: "image_1.jpg", detection-results: "image_1.txt").

Step 3: In these files, each line should be in the following format:

    <class_name> <confidence> <left> <top> <right> <bottom>

Step 4: E.g. "image_1.txt":

    tvmonitor 0.471781 0 13 174 244
    cup 0.414941 274 226 301 265
    book 0.460851 429 219 528 247
    chair 0.292345 0 199 88 436
    book 0.269833 433 260 506 336

Step 5: Put all the files in the folder YOLOX-Bees/map/input/DATASET_NAME/detection-results, where DATASET_NAME can be for example Chueried_Hive01.

To know all the datasets name you can refer to lines 30 - 41 of the file evaluation.py.

At the end the folder YOLOX-Bees/map/input/ should have the following structure:

input
   |——————Chueried_Hive01
   |        └——————detection-results
   |        
   |——————ClemensRed
   |        └——————detection-results
   | 
   |——————Doettingen_Hive1
   |        └——————detection-results
   | 
   |——————Echolinde
   |        └——————detection-results
   | 
   |——————Erlen_diago
   |        └——————detection-results
   | 
   |——————Erlen_front
   |        └——————detection-results
   | 
   |——————Erlen_Hive11
   |        └——————detection-results
   | 
   |——————Erlen_smart
   |        └——————detection-results
   | 
   |——————Froh14
   |        └——————detection-results
   | 
   |——————Froh23
   |        └——————detection-results
   | 
   |——————Hempbox
   |        └——————detection-results
   | 
   |——————UnitedQueens
   |        └——————detection-results

Evaluation

Step 1: Download from Azure the zip file /beelivingsensor/dslab2021/dslab2020_bee_detection_data_blurred/reproduce_results_dataset.zip, unzip it and put all the folders inside the folder YOLOX_Bees/datasets/.(we need them to create ground truth labels).

Step 2: Run the following command to obtain predictions for all the datasets

python evaluation_no_yolox.py image -f exps/default/yolox_bees_eval.py  --tsize 832

Step 3: At the end you will find a file called mAP_results.txt together with an output file for each dataset in the folder YOLOX-Bees/map/output/.

How to setup Euler

Step 1: To login open the terminal and run the following command (you must use ETH VPN):

 ssh ETH_USERNAME@euler.ethz.ch

Step 2: Run the following command in order to install modules:

env2lmod
module load eth_proxy gcc/6.3.0 python_gpu/3.8.5

Step 3: To move files from your PC to Euler or vice-versa you can use scp command.

Step 4: Go to the section How to reproduce results and follow the instructions.

Euler commands

1 bpeek -> Use bpeek to check the output of the job you are interested in:

bpeek jobID

2 bkill -> Use bkill to kill a specific job:

bkill jobID

3 bjobs -> Use bjobs to check all your active/pending jobs:

bjobs 

4 vim log_test -> Use this command to check the log/errors of a completed/interrupted job:

vim log_test

Deployment

1. MegEngine in C++ and Python
2. ONNX export and an ONNXRuntime
3. TensorRT in C++ and Python
4. ncnn in C++ and Java
5. OpenVINO in C++ and Python

Third-party resources

• The ncnn android app with video support: ncnn-android-yolox from FeiGeChuanShu
• YOLOX with Tengine support: Tengine from BUG1989
• YOLOX + ROS2 Foxy: YOLOX-ROS from Ar-Ray
• YOLOX Deploy DeepStream: YOLOX-deepstream from nanmi
• YOLOX MNN/TNN/ONNXRuntime: YOLOX-MNN、YOLOX-TNN and YOLOX-ONNXRuntime C++ from DefTruth
• Converting darknet or yolov5 datasets to COCO format for YOLOX: YOLO2COCO from Daniel

Cite YOLOX

If you use YOLOX in your research, please cite our work by using the following BibTeX entry:

 @article{yolox2021,
  title={YOLOX: Exceeding YOLO Series in 2021},
  author={Ge, Zheng and Liu, Songtao and Wang, Feng and Li, Zeming and Sun, Jian},
  journal={arXiv preprint arXiv:2107.08430},
  year={2021}
}