Coonamessett Annotation

[aus10powell]

Create Golden Standard Dataset for Coonamessett Fishery Site

Background

We need to establish a golden standard dataset for the Coonamessett fishery site using data from the 2024 season. This dataset will reflect the current camera conditions and latest model training methods, serving as a benchmark for future models.

Objective

Create a high-quality, manually annotated dataset from the 2024 season videos, focusing on scenes with multiple fish to address label imbalance issues.

Requirements

1. Use video file: 1_2024-05-14_09-00-00_987
2. Create a spreadsheet with the following columns:
   • "Minute": The minute mark since the video started
   • "Count": Number of fish that have passed the mid-point of the screen within the minute window

Example Spreadsheet Format:

Minute	Count
1	0
2	0
3	0
4	0
5	0
6	25

Note: This example shows a sudden increase in fish count at minute 6, which is a good representation of the label imbalance issue we're addressing.

Process

1. Download the video file from Dropbox
2. Use a video player to carefully review the footage
3. Count fish passing the mid-point of the screen for each minute
4. Record counts in the spreadsheet following the format shown above

Reference

For comparison, the video file 2_2018-05-10_06-39-30 from the IRWA dataset contains 4 fish in a much shorter duration.

Potential Challenges

• Accurate counting may be difficult when fish are swimming quickly or overlapping in multiple frames
• Maintaining focus during long periods without fish activity
• Handling sudden spikes in fish count, as seen in the example (0 to 25)

Deliverables

1. Completed spreadsheet with minute-by-minute fish counts for the entire video duration
2. Brief report on the annotation process, including any challenges encountered or observations made

Additional Notes

• Ensure consistent counting methodology throughout the video
• If possible, have multiple team members validate the counts to ensure accuracy
• Pay special attention to minutes with high fish counts to ensure accuracy

Please comment on this issue if you need any clarifications or have suggestions for improving the annotation process.

[TalhaNadeem001]

Hi aus10powell

I hope you're doing well. I was reviewing your project and noticed that I am unable to locate the video file named 1_2024-05-14_09-00-00_987 in your Dropbox. Could you please check if the file has been uploaded correctly or provide an alternative link if available?

Thanks in advance

[aus10powell]

Hi Talha,

Thank you for checking. Are you a part of MIT? This file is currently a student effort so would have to check if you are not

[TalhaNadeem001]

I apologize for the oversight. The project caught my attention, and I was eager to learn more about it unfortunately i am not part of MIT

[aus10powell]

@TalhaNadeem001 not a problem. Not obvious from README. I can check with the group and see if videos are available for open-source community

[Zhongqi-Chen]

40 video clips were chosen based on the 2024 stream-side count data, 30 have been counted by two volunteer counters. Each video is 10 minutes long.

Example format:

Time	Frame	Species	Direction
00:07:54	7108	Herring	up
00:07:55	7119	Herring	down