PFTrack Support Forum

I feel like I've run down a rabbit hole with how PFTrack handles OBJs with textures. I've been slowly adding my own custom primitives and objects for image modeling or checking my solution. I'd worked through how to get .obj's from my right handed software to PFTrack's left handed space, but I could never figure out why my textures never rendered (really hadn't spent the time to investigate until now). I finally decided to dig in on this and discovered that PFTrack only reads the texture from the MTL if a relative path is used. I totally understand that relative paths make moving OBJ's more portable, but I was suprised that absolute paths don't work. Is there a reason for this, seems like either should be able to be handled? I also noticed that I get differnt behavior for rendering the OBJs I create depending on the node I add them to. When I create my OBJs for use with PFTrack I have to flip the normals to get them to render properly in the Image Modeling node but with one fo these when I copied it to the Objects directory and add it to the node it looks like it is flipped inward. I know the Test Object node has a coordinate system parameter, but the Image Modeling does not. This should work right, the same OBJ in both Image Modeling and Test Objects? Lastly, with Test object node, when I set the display render to Texture, I can't the texture from the MTL to render, it's a copy of the same object I'm using in the primitives node, so I would expect it to work. Are there details, I'm missing here or are there some bugs here?

Is there a bug where you can't have multiples of the same node type (cameraSolver) within a Group in a Saved Tree preset? i save a Tree Preset with 3 camera solvers in the tree. i do this for a Python script that solves the 3 camera Translation types simultaneously so i can then choose the best result. the problem is the Tree Preset saver keeps popping 2 of the camera solver nodes outside of the Group when i add the saved group back into my current node page. it's annoying since adding them back in to the group manually reverts the group back to the default name & color. i checked the .XML and all of the nodes are included in the 'nodeId' listing so should be included in the group. the XML looks like it should work & keep everything together.

#postshot #exportscripts This is an experimental export script for exporting cameras and points from PFTrack 24.12.19 and later to Jawset Postshot for Gaussian Splat training https://www.jawset.com/ Before using the script, please review the usage guidelines below to get the best results. Download zip file Shot Setup The script can export movie or photogrammetry cameras, along with tracking points and point clouds. Download and unzip the file into your Documents/The Pixel Farm/PFTrack/exports folder and relaunch PFTrack. This will create a new export format in the Scene Export node called "Jawset Postshot (.json)" When setting up your cameras in PFTrack, it is important to ensure you are correcting for lens distortion. Gaussian Splatting is initialised from your point dataset and trains a radiance field to match your image data, so the results you get will depend strongly on the quality of your input cameras and points. A sparse set of points may not initialise the training as well as a dense point cloud, and datasets with low parallax or coverage may not give the best results when viewed from angles other than your original cameras, so please refer to the the Postshot user guide for capturing guidelines: https://www.jawset.com/docs/d/Postshot+User+Guide Point density You should make sure your shot has enough tracking points to initialise the Postshot training. If you've just used a few User Track points or a small number of Auto Track points, you will probably get better results by adding some more to your shot. You can do this easily by placing an empty Auto Track node upstream from your Camera Solver before solving. Then, after you've solved your camera and are happy with the result, go back to your Auto Track node and generate more tracking points, increasing the Target Number up to 500 or more. In your camera solver, select all your tracking points and click the Solve Trackers button to solve for their 3D positions whilst keeping the camera fixed. Alternatively, you can use the Select Frames node to decimate your movie clip into a set of photos, and then use the Photo Cloud node to create a dense point cloud, and attach both the camera and dense point cloud to the export node as shown here: Tree layout for adding a dense point cloud The Select Frames node could also be used before exporting your movie camera to reduce the number of frames being loaded into Postshot if you have a very long image sequence. Exporting from PFTrack Select the "Jawset Postshot (.json)" export format and make sure to enable the "Undistorted clip" Distortion export option, setting the image format to either JPEG, TIFF or OpenEXR with suitable frame number padding. We recommend using TIFF or OpenEXR as this will ensure invalid pixels around the boundary of your undistorted images are written with a zero in the alpha channel and ignored by Postshot. Alternatively, make sure your undistorted images are cropped to the original image size during the solve to reduce empty pixels as much as possible. After exporting, you will find a .json file and a .ply file in the export folder containing your camera and point data respectively, along with your undistorted images in the clips folder. Importing into Postshot You can drag-and-drop the entire export folder directly into Postshot, but it is important to ensure no other files are present in the folder. Macos users in particular should remove all .DS_Store files that are created when opening the folder in Finder, as they will prevent the dataset from loading and give an "invalid string position" error message. In Postshot, make sure to enable the "Treat Zero Alpha as Mask" option to ensure the boundary pixels in the undistorted images are ignored during training. Please refer to the Postshot user guide for all other settings.