Timeline for Object Localisation without Classification
Current License: CC BY-SA 4.0
19 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| S Feb 16, 2019 at 15:00 | history | bounty ended | Swapnil Rustagi | ||
| S Feb 16, 2019 at 15:00 | history | notice removed | Swapnil Rustagi | ||
| Feb 15, 2019 at 6:00 | history | tweeted | twitter.com/StackStats/status/1096287993747460096 | ||
| Feb 14, 2019 at 21:11 | comment | added | EngrStudent | I think this is one of two things. Either it is trivially easy, or you picked an unsolvable problem. The set of all sets has itself as a subset. If you have shoes, do you have aglets? Do you have treads? Do you have threads? There are always sub-features. So what is the "feature of interest". Either the background color is specified, and you are bounding the single non-background object or its not and you are having to say what is the meaningful component. | |
| Feb 14, 2019 at 8:50 | comment | added | Swapnil Rustagi | @user20160 It's not completely simple, as I have tried GrabCut algorithms and Canny edge detection with mean intersection over union coming out to be between 0.5 and 0.6. Few images may not be a good representative of my dataset here, since GrabCut performed better for some of them, and edge detection performed better on others. | |
| Feb 14, 2019 at 8:28 | answer | added | user5302 | timeline score: 2 | |
| Feb 11, 2019 at 21:35 | comment | added | user20160 | How simple is the background--can you say more about it? If it's simple enough, machine learning may not be needed at all. Can you post some example images? | |
| Feb 9, 2019 at 14:02 | answer | added | galoosh33 | timeline score: 2 | |
| S Feb 9, 2019 at 14:01 | history | bounty started | Swapnil Rustagi | ||
| S Feb 9, 2019 at 14:01 | history | notice added | Swapnil Rustagi | Draw attention | |
| Feb 8, 2019 at 11:55 | comment | added | Swapnil Rustagi | @Rickyfox I have edited my post to make it clearer that I do have a training set, but not class labels. | |
| Feb 8, 2019 at 11:54 | history | edited | Swapnil Rustagi | CC BY-SA 4.0 |
Added a tag and clarified one sentence
|
| Feb 8, 2019 at 11:22 | comment | added | Swapnil Rustagi | @Rickyfox I meant I have a training set. But it only contains the coordinates of the bounding box, not the classes of objects detected. | |
| Feb 8, 2019 at 11:08 | comment | added | deemel | Then I must have misunderstood you when you wrote "but neither do I want to classify the objects nor do I have a training set for classification." . | |
| Feb 8, 2019 at 10:34 | comment | added | Swapnil Rustagi | @Rickyfox Thanks a lot for replying. While I may use an unsupervised learning model like a clustering approach, it needs to be noted that the background is not 100% pixel-perfect white (they are photos clicked against a white background). Also I have training data, and I don't think discarding that and using clustering would give me good accuracy. Sorry to reject your edit, but I think completely unsupervised learning algorithms may not be the answer here. | |
| Feb 8, 2019 at 10:14 | comment | added | deemel | Off the top of my head it seems like anomaly detection or a simple clustering approach might be of interest here. Assuming that the single object you want to localize in the image presents one coherent shape (or close to it) that differs from the white background, so the "not purely white" observations (pixels/segments?) would form a cluster that needs to be detected. Note however that I'm fairly unfamiliar with image processing problems. | |
| Feb 8, 2019 at 10:11 | review | Suggested edits | |||
| Feb 8, 2019 at 10:26 | |||||
| Feb 7, 2019 at 12:45 | review | First posts | |||
| Feb 7, 2019 at 17:41 | |||||
| Feb 7, 2019 at 12:43 | history | asked | Swapnil Rustagi | CC BY-SA 4.0 |