Requests: Make ICE better accept its Output as Input // Combine ICE partial solutions // Loosely Structured Panos RRS feed

  • General discussion

  • Several times over the years I have tried to stitch images with ICE which failed when I put all of the photos for the panorama into ICE at the same time.

    In putting portions of the pano into ICE at a time, I've often found that ICE will stitch portions of the pano flawlessly.
    Unfortunately, there appears to be no way to easily combine these partial solutions into a complete whole.

    I know that you often tell people that ICE does not support panoramic images, but it seems like ICE ought to be as intelligent about reading its own output images which have been exported to disk as HD View is.

    Likewise, I've often wished that I could save each partial solution to a layered Photoshop file (thanks for the great feature) but then drag multiple layered Photoshop files into ICE and have ICE do a quick alignment based on the stitched base layer of each PSD file and then fine tune each layer's position into a single arrangement.

    Similarly, I've often wished that I could save each successful partial solution's SPJ file and then drag multiple overlapping SPJ projects into ICE and have it use each individual SPJ's existing solution as a baseline and then adjust the perspective so that things fit together more flawlessly.

    I have shot panoramas on partially cloudy summer days when the clouds are blowing along pretty quickly where I shot the panorama in columns on a tripod and ICE failed every time to stitch the entire pano, but it could stitch each column of photos without a problem. Unfortunately these individual exported columns would not stitch together. I also tried selecting rows through these (although this was a bit difficult since each column didn't pan the same number of degrees between input shots) and these would combine well enough or a square of imagery toward the horizon would work but every attempt to combine these partial shots into a whole would fail.

    It seems to me that having an option that sits between the 100% automated panorama and the perfectly Structured panorama where I can specify ranges of photos as equating to columns or rows, specify the panning direction for each range, and click and drag the columns and rows into a rough alignment would be ideal when everything else is failing to solve for a satisfactory alignment.

    Another common scenario for me is wanting to use every frame of a video to use every available pixel for the edges of the pano.
    By converting the video to an image sequence and then selecting every eighth or tenth or fifteenth or thirtieth image and providing them to ICE, ICE will succeed and it is easy to stitch every sub-selection of images together (for example all ten sets of every tenth photo), but there is not then any way to combine these sub-selections where ICE has already figured out how each of these sub-selections align so that I am able to align every frame of the video to use with ICE's projection options before exporting to a layered PSD for fine tuning and choosing perfect poses for people or doing manual clone brushing to cancel out any parallax errors.

    I'd also really love an option to save the final alignment of all frames as a panoramic video where you can watch the original frames move against the background of all stitched frames.

    I know that many times the reason that ICE fails with the entire set of images is due to non-ideal input in terms of exposure or overlap, but the bottom line is that ultimately I have to work with what has actually been captured, since many times a reshoot is not possible.

    With regards to stitching a pano from video, it seems to me that surely future versions of ICE could follow the relatively simple workflow that I've described above of doing a preliminary pass (as it currently does) with a sparse selection to find how the frames overlap each other (especially in cases where the camera motion is not a constant speed and/or direction) but then combine multiple sparse passes until it is able to utilize all input frames.

    It is very frustrating to me that both ICE and Photosynth currently choke and die when you give the dense coverage that video frames provide (a case that the average user would assume would be easier to solve due to diminished difference). I do understand that this means more image features to hold in RAM and more image features which will definitely match each other, but these technologies have such great potential if they could only scale to video scale.

    Apologies that this ranges over several different topics, but they do relate.

    Thursday, February 26, 2015 10:24 PM