/NOAA_Fish_Detection

NOAA_Fish_Detection

All codes are here: https://microsoft-my.sharepoint.com/:f:/p/antriv/EvhEqYagibZHkx4oe1x7YIgBlu907QDHgSre6E4xaL-uew?e=yYTieS

Please download and extract the zip files.

Prerequisites

Python: 3.5, CUDA: 10.0, CUDNN:7

download the "req.txt" file here

If running for multiple videos, make sure they are all in one directory

Files Needed

  • noaa_imerit_1_main_inference_single_videoUse.py
    • Used to detect fish for one video
  • noaa_imerit_2_main_inference_multiple_videoUse.py
    • Used to detect fish for multiple videos in one directory
  • noaa_imerit_OD_inference_mainUse.py
    • Used to generate a csv of frame numbers(ie timestamps) for when fish are detected for a single video
  • noaa_imerit_main_inference_single_video.py
    • Used to detect fish for one video(SHOULD NOT BE RUN IN TERMINAL)
  • noaa_imerit_framesUse.py
    • Used to generate a folder with images generated by both v1 and v2 models for a single video
  • noaa_imerit_main_condition_detection.py
    • Used to produce two csvs: a copy of the original csv with bounding box ids labeled and another csv with timestamps of bounding box ids

Set Up

  1. conda create -n noaa_env python=3.5 anaconda

  2. conda activate noaa_env

  3. pip install –r req.txt

  4. Run the following command on terminal: export PYTHONPATH=$PYTHONPATH:<user_path>/models/research:<user_path>/models/research/slim

Fish Detection for GUI Single Video:

  1. Install "X"-Server locally (e.g: XMing Server)

  2. Activate "-X" option when connecting to VM

  3. Run the following command on terminal: python <dir_path>/models/research/object_detection/noaa_imerit_OD_inference.py

Fish Detection for Single Video (terminal):

  1. mkdir <dir_path>/Project_NOAA_imerit/outimg

  2. mkdir <dir_path>/Project_NOAA_imerit/outimg/csvfiles

  3. python <dir_path>/models/research/object_detection/noaa_imerit_1_main_inference_single_videoUse.py

    --pathVideo <dir_path>/Project_NOAA_imerit/data/test/videos/2019_GOPR1207_analyze.MP4

    --pathCSV <dir_path>/Project_NOAA_imerit/outimg/csvfiles/

    --pathIMG <dir_path>/Project_NOAA_imerit/outimg/

  4. This will produce one csv file. The last column of the csv file will be set to 0 for all rows as a default.

Fish Detection for Multiple Video(terminal):

  1. mkdir <dir_path>/Project_NOAA_imerit/outimg

  2. mkdir <dir_path>/Project_NOAA_imerit/outimg/csvfiles

  3. python <dir_path>/noaa_imerit_2_main_inference_multiple_videoUse.py

    --pathVideo <dir_path>/Project_NOAA_imerit/data/test/videos/

    --pathCSV <dir_path>/Project_NOAA_imerit/outimg/csvfiles/

    --pathIMG <dir_path>/Project_NOAA_imerit/outimg/

  4. This will produce a csv file for each video. The last column of each csv file will be set to 0 for all rows as a default.

Labeling Bounding Box IDs

  1. For this step, make sure that a csv is created from either running fish detection for multiple or single video

  2. python noaa_imerit_main_condition_detection.py

    --pathCSV <dir_path>/Project_NOAA_imerit/outimg/csvfiles/<csv_name> --threshold

  3. Step 2 creates two csvs. 1) An identical csv to the input one, witth bounding boxes all given an id. 2) A csv with end & start timestamps of bounding boxes

Output Expected

  • Three CSV files for each video file in the CSV directory

    • Original csv file has the naming convention:

    • Bounding Box ID csv file has the naming convention: __condition_detection.csv

    • The csvs have 10 columns: video_name, frame_number, timestamp, number_boxes, index, x_min, y_min, x_max, y_max, box_id(is useful after running noaa_imerit_main_condition_detection.py)

    • Key for each csv column(same for both csv files)

      video_name: name of video

      frame_number: frame of the video

      timestamp: time in video in which frame begins

      number_boxes: number of bounding boxes in the frame_number

      index: ith bounding box in frame_number

      x_min: x-coordinate of bottom left corner of the bounding box

      y_min: y-coordinate of bottom left corner of the bounding box

      x_max: x-coordinate of upper right corner of the bounding box

      y_max: y-coordinate of upper right corner of the bounding box

      box_id: an id given to the bounding box(initially 0, changes after running noaa_imerit_main_condition_detection.py)

    • Bounding Box Timestamp csv file has naming convention: __bounding_box_timestamp.csv

    • The csv has 5 columns: video_name, bounding_box_id, start_time, end_time, duration

    • Key for each csv column:

      video_name: name of video

      bounding_box_id: id of the bounding box

      start_time: time of first appearance of bounding box

      end_time: time of last appearance of bounding box

      duration: duration of bounding box within the video

  • One IMAGE folder for each video in the IMAGE directory

    • Each image folder has the naming convention:
    • Each image within each folder has the naming convention:
    • Inside IMAGE folder will be images generated by both v1 and v2 models for a single video