Adding readme, fix test_oop

README.md

@@ -0,0 +1,162 @@
+# YoloV7 | DeepSort and Count
+
+**#ComputerVision** **#RealTimeDetection** **#RealTimeCounting** **#DeepLearning** **#DeepSort** **#MachineLearning** **#AI** **#OpenCV** **#Python** **#YOLO** **#MOT #Docker** **#DigitalImageProcessing**
+
+![Test Cars](readme-img/messi.gif)
+
+## Principal idea
+
+This project started with the idea of being able **to help farm workers who still count cattle by hand.**
+
+Starting with this idea, for a project on Digital Image Processing at the [National University of Río Cuarto](https://www.unrc.edu.ar/), my research began to achieve the first objective.
+
+As the project developed, I realized that I could further generalize its scope. It wouldn't just be cattle, but could be any class, as long as we can train the detection model.
+
+I have managed to **make detections in "real time" (almost 30 FPS)** with videos or images from different types of sources.
+
+In addition to the detections, **it has been possible to track these objects and detect when they enter a certain region of interest (ROI)** defined by the user **and make a count by class independently** detected, for example:
+
+- person: 2; car: 5.
+
+### Information flow
+
+![Information flow](readme-img/Untitled.png)
+
+### Count algorithm
+
+![Count Algorithm](readme-img/Untitled%201.png)
+
+## Results
+
+- One class
+    - Source: YouTube
+    - Invert Frame: False
+    - Classes: car → [0]
+
+
+![Test Cars](readme-img/test_cars.gif)
+
+
+### Test environment
+
+I’ve done test with:
+
+- **PC:** HP-OMEN ce004-la
+- **GPU:** NVIDIA 1060 MAX-Q DESIGN 6GB
+- **CPU:** INTEL I7 7700HQ
+- **RAM:** 16GB
+- **DISK:** SSD M2
+
+## Performance
+
+- I’ve obtained almost **30 FPS (detection + tracking time**) in the majority of the tests.
+- Performance suffers when the number of concurrent detections per frame rises above 10/20. We can suffer a drop in fps (detection + tracking) to about 5 per second.
+
+
+## Problems
+
+- You need a lot of processing capacity.
+- Occlusion it is a problem (when the objects are too close)
+- When an object get out of the frame we lost the tracking and then the program consider a new object if this come into the frame again.
+
+
+## Ideas for the future
+
+- You can create an **API.**
+- You can **re-train the detection model with your own class/classes** and run the program.
+- You can add an **visual interface.**
+- You can **collaborate on this repository** integrating new features or solving bugs!
+- 
+
+---
+## Try it yourself. Do a test!
+
+### Warnings
+
+- **NVIDIA Graphic Card** needed to run this project.
+- **Docker** previously installed in your computer.
+- **NVIDIA Docker image** needed.
+
+### Configuring your HOST PC
+
+Do this following steps *in order.*
+
+**Install Docker**
+
+1. [Install Docker](https://docs.docker.com/engine/install/ubuntu/)
+
+2. [Post-Install](https://docs.docker.com/engine/install/linux-postinstall/)
+
+**Install Nvidia Drivers**
+
+- Instalation:
+
+    ```bash
+    ubuntu-drivers devices # To know which driver is recommended
+    sudo ubuntu-drivers autoinstall # To automatically install the recommended driver
+    reboot # We need to reboot the system after the installation
+    ```
+
+
+- Checks:
+
+    ```bash
+    nvidia-smi # Command to check if the driver was installed correctly: The output must be a list of GPU's and a list processes running on it
+    sudo apt install nvtop # Program to check the GPU usage
+    nvtop # To run the nvtop
+    ```
+
+    ![nvidia-smi](/readme-img/Untitled%202.png)
+
+**Pull nvidia-gpu image**
+
+Follow this  **[installation guide.](https://docs.nvidia.com/datacenter/cloud-native/container-toolkit/install-guide.html)**
+
+**Pull this repository**
+
+```bash
+git pull https://github.com/aagustinconti/yolov7_counting
+```
+
+**Change permissions of ./start.sh**
+
+```bash
+sudo chmod +x ./start.sh
+```
+
+**Download YOLOv7x pretrained weights**
+
+1. Download the pretrained weights of YOLOv7x [here.](https://github.com/WongKinYiu/yolov7/blob/main/README.md#performance)
+2. Save them in **./pretrained_weights** directory.
+
+**RUN ./start.sh**
+
+To build the docker image if is not available in your PC and run the container with the propp
+
+```bash
+run ./start.sh
+```
+
+### Configuring the test
+
+- Into the **test_oop.py** file you can modify some characteristics of the instance of the class *YoloSortCount()* before to execute the *run()* method.
+    - By default:
+        - Source: WebCamera (0)
+        - Invert frame: True
+        - Classes: person (ID: 0)
+        - Save: False
+
+**RUN the test**
+
+On the docker terminal run
+
+```bash
+ python test_oop.py
+```
+
+
+## Thanks to
+
+[YOLOv7](https://github.com/WongKinYiu/yolov7/)
+
+[SORT Algorithm](https://github.com/dongdv95/yolov5/blob/master/Yolov5_DeepSort_Pytorch/track.py)

start.sh

@@ -1,32 +1,5 @@
 #!/bin/bash
 
-######## Host PC ########
-
-# Pre run
-# 1. sudo chmod +x start.sh
-
-# Do the complete instalation of Docker.
-
-# 1. Install: https://docs.docker.com/engine/install/ubuntu/
-# 2. Post-install: https://docs.docker.com/engine/install/linux-postinstall/
-
-
-# Install Nvidia Drivers.
-
-# Instalation:
-# 1. ubuntu-drivers devices  (To know which driver is recommended)
-# 2. sudo ubuntu-drivers autoinstall (To automatically install the recommended driver)
-# 3. reboot # We need to reboot the system after the installation
-# Checks:
-# 1. nvidia-smi (Command to check if the driver was installed correctly: The output must be a list of GPU's and a list processes running on it)
-# 2. sudo apt install nvtop (Program to check the GPU usage)
-# 3. nvtop (To run the nvtop)
-
-# Pull nvidia-gpu image
-
-# 1. Instalation guide: https://docs.nvidia.com/datacenter/cloud-native/container-toolkit/install-guide.html
-
-
 # Create the image of torch
 
 echo "Checking if the Image already exists..."

test_oop.py

@@ -1,39 +1,141 @@
 from yolov7_sort_count_oop import YoloSortCount
 
+#################### TEST #################### 
+
+# INSTANCIATE
 
-# Test
 test = YoloSortCount()
 
-# Source
-test.video_path = 0#"https://www.youtube.com/watch?v=qP1y7Tdab7Y" | "http://IP/hls/stream_src.m3u8" | 0 | "img_bank/cows_for_sale.mp4"
 
-test.max_fps = 1000 #Max 1000
+
+"""
+###### AVAILABLE SOURCES ###### 
+WebCamera: 0 ---> DEFAULT
+Youtube Video or stream: "https://www.youtube.com/watch?v=qP1y7Tdab7Y"
+Stream URL:  "http://IP/hls/stream_src.m3u8"
+RSTP Stream: 
+Local video: "img_bank/cows_for_sale.mp4"
+Local image: "img_bank/img.jpg" | "img_bank/img.png" 
+"""
+test.video_path = "https://www.youtube.com/watch?v=emI8r2dfk6g"
+
+
+"""
+###### FRAME PROPERTIES ###### 
+
+- Set the max size of the frame (width)
+- Set the max fps of the video
+- Invert the image (In case of your WebCamera is mirrored, IE)	
+"""
 test.max_width = 720
+test.max_fps = 25 #Max 1000
+test.inv_h_frame = False
+
 
-# Show results
+
+"""
+###### SHOWING RESULTS ###### 
+
+- Show the results in your display (Interactive ROI, imshow of the out frame)
+- In case of you are not showing the results, set the timer to stop the execution.
+- Stop the frame with hold_image method in case you are using image as a source.	
+"""
 test.show_img = True
+test.ends_in_sec = 10
 test.hold_img = False
 
+
+"""
+###### ROI ###### 
+
+- Load the ROI manually.
+- 
+- Load the ROI color.	
+"""
+#test.roi = [0,0,0,0]
 test.auto_load_roi = True
+test.roi_color = (255, 255, 255)
 
-test.ends_in_sec = 10
 
-# Debug
+
+"""
+###### DETECTION MODEL ###### 
+
+- Specify the path of the model.
+- Select the ID of your Graphic Card (nvidia-smi)
+- Select the classes to detect	
+- Set the image size (Check if the YOLO model allows that --> IE: yolov7.pt 640, yolov7-w6.pt 1280 or 640)
+- Set the bounding box color
+- Set the minimum confidence to detect.
+- Set the minimum overlap of a predicted versus actual bounding box for an object.
+"""
+test.model_path = 'pretrained_weights/yolov7.pt'
+test.graphic_card = 0
+test.class_ids = [0,2]
+test.img_sz = 640
+test.color = (0, 255, 0)
+test.conf_thres = 0.5
+test.iou_thres = 0.65
+
+"""
+###### TRACKING MODEL ###### 
+
+- Specify the path of the model.
+- Set the max distance between two points to consider a tracking object.
+- Set the max overlap of a predicted versus actual bounding box for an object.
+- Set the image size (Check if the YOLO model allows that --> IE: yolov7.pt 640, yolov7-w6.pt 1280 or 640)
+- Set max_age to consider a lost of a tracking object that get out of the seen area.
+- Set the minimum frames to start to track objects.
+- Set the value that indicates how many previous frames of feature vectors should be retained for distance calculation for each track.
+- Set the color of the centroid and label of a tracking object.
+"""
+test.deep_sort_model = "osnet_x1_0"
+test.ds_max_dist = 0.1
+test.ds_max_iou_distance = 0.7
+test.ds_max_age = 30
+test.ds_n_init = 3
+test.ds_nn_budget = 100
+test.ds_color = (0, 0, 255)
+
+"""
+###### PLOT RESULTS ###### 
+
+- Specify the min x (left to right) to plot the draws
+- Specify the min y (top to bottom) to plot the draws
+- Specify padding between rectangles and text
+- Specify the text color.
+- Specify the rectangles color.
+"""
+test.plot_xmin = 10
+test.plot_ymin = 10
+test.plot_padding = 2
+test.plot_text_color = (255, 255, 255)
+test.plot_bgr_color = (0, 0, 0)
+
+
+
+"""
+###### DEBUG TEXT ###### 
+
+- Show the configs
+- Show the detection output variables
+- Show the tracking output variables
+- Show counting output variables
+"""
 test.show_configs = False
 test.show_detection = False
 test.show_tracking = False
 test.show_count = False
 
-# Detection model
-test.class_ids = [0]
-test.conf_thres = 0.5
+"""
+###### SAVING RESULTS ###### 
 
-# Frame
-test.inv_h_frame = True
+- Select if you want to save the results
+- Select a location to save the results
+"""
+test.save_vid = True
+test.save_loc = "results/messi"
 
-# Save
-test.save_loc = "results/test_test"
-test.save_vid = False
 
 # Run
 test.run()

yolov7_sort_count_oop.py

@@ -1,4 +1,4 @@
-# https://learnopencv.com/yolov7-object-detection-paper-explanation-and-inference/
+# https://github.com/WongKinYiu/yolov7/
 
 # Pytorch
 import torch