Using Timelapse AND Projects for same data set

Workflow
1

Hi - is it possible to use both “Analyse a folder” (for use in Timelapse), as well as “Build a project” on the same data without having to run both processes separately?

Timelapse has some features that users are familiar with and integrated with other external reporting/analysis solutions, but AddaxAI Projects have very useful mapping, activity and other overview info that is not found in TimeLapse.

“A project does not make a recognition file for Timelapse, only “Analyse a folder” does that” - but is there a way to generate a recognition file for Timelapse in any way when data is processed in a Project, even if just once off before further analysis or verification, especially when deployments are analysed one per folder in a project?

2

Hi @Heinrich,

Good question, but the answer is no, a project cannot make a recognition file. Let me explain the thinking behind it.

  1. Timelapse and projects organise data differently. Timelapse works on one root folder: all images sit below it, and the recognition file describes them by their path relative to that folder. A project doesn’t think in folders like that. Its deployments can be scattered all over your device, some on external drives, and they can move. Those two don’t match, so a recognition file made from a project would work for some deployments and break for others.

  2. There is also a workflow reason. The recognition file is a hand-off. It says: here are the AI results, I continue in Timelapse. From there the verification and further analysis happen in Timelapse or another tool, and nothing comes back, there is no re-import into AddaxAI. A project is the opposite: it’s the workflow where AddaxAI does the interpreting, with verification, counts, maps and charts inside. If a project would also export to Timelapse, you would end up with two verification records of the same data that can never be synced, and it becomes unclear which one is the truth.

So it’s meant as a choice: “Analyse a folder” if you continue in Timelapse, “Build a project” if you continue in AddaxAI. The AI results are identical either way.

One thing that may help for a single folder: run “Analyse a folder” with the recognition file on, and promote the run to a project afterwards. Nothing is analysed again and your corrections carry over to the new project. The project just won’t update the JSON later. How that works is described here:

Hope this helps! Cheers,

Peter

3

Thanks for clarifying!

Another relaterd question, based on promoting a run to a project: “run “Analyse a folder” with the recognition file on, and promote the run to a project afterwards” - is a new project created during this “promote” step, or can the run be associated with an existing project?

And it doesn’t seem as if projects can be combined, if multiple runs were promoted to projects afterwards - is this somewhere on the development roadmap? :slight_smile:

4

Hi @Heinrich,

Good questions again. Promote always makes a new project. It takes the run itself and turns it into a project, with all detections and corrections in it. So it can’t be added to an existing project.

And no, projects can’t be combined yet. Moving a deployment from one project to another is not built.

But for several folders there is an easy way. Build the project first, add all folders as deployments, and run once. The AI results are the same as in a folder run, so you lose nothing.

Before I think about merging, I’m curious about your use case. Why would you like the same data in Timelapse and in a project? I ask because those two don’t go together well. Timelapse is a hand-off, you verify and analyse there and nothing comes back. A project is the opposite, you verify inside AddaxAI. If you do both, you get two verification records of the same images that slowly drift apart.

Cheers,

Peter

5

Hi Peter

I think it boils down to AI first vs Human first approach differences between AddaxAI and Timelapse.

In my opinion Timelapse was designed as a tool for humans to visually inspect images and videos, combine them into anotated observations with counts and classifications, and produce results that can then be exported for further processing and analysis. It’s an exceptional tool for managing images and associated data - the ability to read from EXIF and add multiple custom fields via the template builder approach is extremely usefull. When AddaxAI (before v7) came along and improved the AI detection and classification process it made the overall process even better.

I currently see AddaxAI as one of the best, if not the best, detection and species recognition solutions for scitizen scientist or individual person research efforts. Being completely offline reduces a lot of data security issues, not to mention upload/download bandwidth in areas where connectivity is either absent, very slow, or very expensive. The ability to use multiple models based on area is very useful.

The areas where I see Timelapse as still being superior is the ability to add additional user defined fields and it’s image-centric inspection interface: the ability to adjust image brightness/contrast and zooming to make visual identification/verification in low light images easier, and to see all detections and classifications in a single image to verify visually is the areas I most miss in AddaxAI.

AddaxAI seems to be primarily designed to automate detection (animal vs human, vehicle or empty) and species recognition using AI - and it does it well. v7+ started moving into user interface and process improvements, and the dashboard-oriented interface with some analytics and especially maps is very useful - thing that needs to be done outside of TimeLapse. But the major research and data management constraint is the inability to add user defined fields to observations. And the inability to easily scroll through images to verify all detections and classifications might be just a personal old school human based process that I’m familiar with and enjoy, but it could improve AddaxAI usability if implemented.

Bottom line is that TimeLapse is a great solution very strong in data management and visual/human processes, and AddaxAI is technically superior with automated AI detection and classification, and seems to be moving into areas overlapping with Timelapse, but not doing it as well (yet). AI is useful to improve and speed up processes, but manual human review and veification processes is still needed and should be user friendly and easy.

My current need to have data in both systems is related to how smaller projects have different needs - a quick and dirty manual review of AI identified species, with added metadata and user defined fields is currentyl the Timelapse sweet spot, and AddaxAI’s bulk detection and classification, with some built in site and deployment management, analytics and mapping makes working with large datasets doable for single researchers.

I think this shoul dclarify my needs and comments - please keep up the great work, and let me know if you have any further questions!

I’ll reserve comments on AddaxAI Connect for a separate discussion, and only comment for now is that I’m exited to start playing iwth it soon :slight_smile:

1 Like