Using Google Vertex AI, I trained a model to detect some specific objects in images.
Can i use this trained model to detect sames objects, but in videos ?
The following requirements apply to datasets used to train AutoML or custom-trained models;
Vertex AI supports the following video formats for training your model or requesting a prediction (annotating a video).
.MOV
.MPEG4
.MP4
.AVI
To view the video content in the web console or to annotate a video, the video must be in a format that your browser natively supports. Since not all browsers handle .MOV or .AVI content natively, the recommendation is to use either the .MPEG4 or .MP4 video format.
Maximum file size is 50GB (up to 3 hours in duration). Individual video files with malformed or empty timestamps in the container are not supported.
The maximum number of labels in each dataset is currently limited to 1,000.
You may assign "ML_USE" labels to the videos in the import files. At training time, you may choose to use those labels to split the videos and their corresponding annotations into "training" or "test" sets.
For best practices on video data used to train your models, you can see this link.
Related
I recently collected video data where the video was generated as image sequences. However, between different video of the same length, different numbers of frames were acquired, which made me think that the image sequence have varied frame rates between videos. So my question is how do I convert this image sequence back to video with accurate duration between frames. Is there a way to get that information from the date and time it was created using a code? I know ffmpeg seems to be the tools many people use.
I am not sure where to start. I am not very familiar with coding, so already have trouble executing the correct codes.
I'm working on a software project in which I have to compare a set of 'input' images against another 'source' set of images and find out if there is a match between any of them. The source images cannot be edited/modified in any way; the input images can be scaled/cropped in order to find a match. The images can be in BMP,JPEG,GIF,PNG,TIFF of any dimensions.
A constraint: I'm not allowed to use any external libraries. ImageMagick is an exception and can be used.
I intend to use Java/Python. The software is purely command-line based.
I was reading on SO and some common image comparing algorithms. I'm planning to take 2 approaches.
1. I could use Histograms/buckets to find out the RGB values of the 2 images being compared.
2. Use SIFT/SURF to fin keypoint descriptors and find the euclidean distance between them and output the result based on the resultant distance.
The 2 images in comparison can be in different formats. An intuitive thought is that before analysis/comparison, the 2 images must be converted to a common format.I reasoned that the image should be converted to the one with lesser quality e.g. if the 2 input images are BMP and JPEG, convert the BMP to JPEG. This can be thought of as a pre-processing step.
My question:
Is image conversion to a common format required? Can 2 images of different formats be compared? IF they have to be converted before comparison, is my assumption of comparing from higher quality(BMP) to lower(JPEG) correct? It'd also be helpful if someone can suggest some algorithms for image conversion.
EDIT
A match is said to be found if the pattern image is found in the source image.
Say for example the source image consists of a football field with one player. If the pattern image contains the player EXACTLY as he is in the source image, then its a match.
No, conversion to a common format on disk is not required, and likely not helpful. If you extract feature descriptors from an image (SIFT/SURF, for example), it matters much less how the original images were stored on disk. The feature descriptors should be invariant to small compression artifacts.
A bit more...
Suppose you have a BMP that is an image of object X in your source dataset.
Then, in your input/query dataset, you have another image of object X, but it has been saved as a JPEG.
You have no idea how what noise was introduced in the encoding process that produced either of these images. There is lighting differences, atmospheric effects, lens effects, sensor noise, tone-mapping, gammut-mapping. Some of these vary from image to image, others vary from camera to camera. All this is done before the image even gets saved to storage in the camera. Yes, there are also JPEG compression artifacts, but to assume the BMP is "higher" quality and then degrade it through JPEG compression will not help. Perhaps the BMP has even gone through JPEG compression before being saved as a BMP.
I'm trying to split a video by detecting the presence of a marker (an image) in the frames. I've gone over the documentation and I see removelogo but not detectlogo.
Does anyone know how this could be achieved? I know what the logo is and the region it will be on.
I'm thinking I can extract all frames to png's and then analyse them one by one (or n by n) but it might be a lengthy process...
Any pointers?
ffmpeg doesn't have any such ability natively. The delogo filter simply works by taking a rectangular region in its parameters and interpolating that region based on its surroundings. It doesn't care what the region contained previously; it'll fill in the region regardless of what it previously contained.
If you need to detect the presence of a logo, that's a totally different task. You'll need to create it yourself; if you're serious about this, I'd recommend that you start familiarizing yourself with the ffmpeg filter API and get ready to get your hands dirty. If the logo has a distinctive color, that might be a good way to detect it.
Since what you're after is probably going to just be outputting information on which frames contain (or don't contain) the logo, one filter to look at as a model will be the blackframe filter (which searches for all-black frames).
You can write a detect-logo module, Decode the video(YUV 420P FORMAT), feed the raw frame to this module, Do a SAD(Sum of Absolute Difference) on the region where you expect a logo,if SAD is negligible its a match, record the frame number. You can split the videos at these frames.
SAD is done only on Y(luma) frames. To save processing you can scale the video to a lower resolution before decoding it.
I have successfully detect logo using a rpi and coral ai accelerator in conjunction with ffmeg to to extract the jpegs. Crop the image to just the logo then apply to your trained model. Even then you will need to sample a minute or so of video to determine the actual logos identity.
Using ffmpeg I can take a number of still images and turn them into a video. I would like to do this to decrease the total size of all my timelapse photos. But I would also like to extract the still images for use at a later date.
In order to use this method:
- I will need to correlate the original still image against a frame number in the video.
- And I will need to extract a thumbnail of a given frame number in a
video.
But before I go down this rabbit hole, I want to know if the requirements are possible using ffmpeg, and if so any hints on how to accomplish the task.
note: The still images are timelapse from a single camera over a day, so temporal compression will be measurable compared to a stack of jpegs.
When you use ffmpeg to create a video from a sequence of images, the images aren't affected in any way. You should still be able to use them for what you're trying to do, unless I'm misunderstanding your question.
Edit: You can use ffmpeg to create images from an existing video. I'm not sure how well it will work for your purposes, but the images are pretty high quality, if not the same as the originals. You'd have to play around with it to make sure the extracted images are exactly the same as the input images as far as sequential order and naming, but if you take fps into account, it should work.
The command to do this (from the ffmpeg documentation) is as follows:
ffmpeg -i movie.mpg movie%d.jpg
I'm writing an interface for a hardware JPEG decoder, and I'm looking for some test images.
Prior to hardware decompression, the software front-end must parse the JFIF and/or EXIF data to obtain the image dimensions and thumbnail. In testing, I found that my current version works well with images obtained from contemporary digital point-n-shoot cameras. In general, the parser obtains the dimensions from the SOF segment, and the thumbnail is retrieved from the EXIF data (if the thumbnail exists).
I'm looking for a broader range of test images to evaluate the system more exhaustively. For example, I have been unable to find any JPEG images that encode the thumbnail in a JFXX (i.e., second APP0) marker. Furthermore, I would like to test the code on a wide variety of images (sizes, progressive scans, etc.). This code is destined for a specialized consumer product, and images are expected to be obtained from a range of digital cameras, both old and new.
Suggestions?
You could get a Flickr API key!