Package is fully functional.
6/30/2020
vv3.1.5 - Now that I have a computer that can hack it, this project seems saved. I have now made sure that the percentages are right. If you're worried about not getting better percentages, it shouldn't bother you. But if you hack this, goto class PNG and change the percentages in the function image_contrast(). You'll have to do some analyzing for your specific data, but the work is still right. So happy hacking!
1/7/2020
v3.0.1 - Missed a journal again. Sorry. Lost my computer. Happy New Year! Anyway, God bless you all. This is going to be one of the final updates to this repo. In order to categorize the information, you will need to set a Tier object's public member variable 'cat' before you send it to ->find_tier(). This will make the directory. Name it right because it will show in the results. Thank you all!
12/27/2019
v2.4.2 - After releasing this morning, I decided to watch a video on CNN and NNs in general. Well i got to work after watching and noticing some things. The apparent reason this wasn't the best example of a NN is because I didn't have a hidden layer. Well, now, though that was debatable, the introduction of the serial dataset was tokenized an invented by myself. So, now, we have an even newer way to see the things we are looking for. 15 times faster though. Muahaha!!
v2.4-patch-1 - patch for saving object file and ignoring duplicate returns.
12/27/2019 (4am)
v2.4 - We're gunning for a releasable product now. Now when it's me saying we, I mean I. I am getting this produced the best I can. So far it looks like the speed is great, and the only parts I haven't been able to fix, is the saving and reusing of the data for futures without having to replay all that processing. BTW, the formula is trademark! Yes! That's right, this is trade copyrighted! That means no one else can use it! BTW, my prorietary tag means, this is for sale. It doesn't mean don't clone. S'ok! Use it and make it better! Happy New Year!
12/22/2019 (Early Morning Time: 1:54am)
v2.3.5 - Brightness added. Now the numbers are surely right.
12/21/2019
v2.3 - After lazily putting out a README with no changes talked about, I sit erady to state that finally my numbers are good. It's in full working condition. TODO: - Add what parts of pictures were so obvious to the match. That's about it.
12/20/2019
v2.2 -I programmed and released without journaling, sorry.
12/19/2019
v2.1 - Refactored and refined the search process as well as how I'm saving files. This entails using a mathematical process rather than bit by bit analysis. This fine-tuning has tripled the speed at least. And the answer finally make sense! Go figure! Math is still king!
Later on 12/18/2019
v2.0 - All comparisons are correctly functioning. We are using a 20% threshold with a max of 4 outputs. Loving it. Works like a dream. Speed is a big problem though. Looking into optimizations is a TODO. I will be looking at putting in the SHA1 Checksum back into the production so that finding originals is simple. Thanks!
12/18/2019
v1.8 - New release allows for more than one possible file to be shown for a given dataset. All necessary functions refactored for such use. Also the match threshold is reduced to 50%.
-patch-2
Requirements changed for sake of operational ease.
(Changed parameters to a function or 2).
v1.7 - Takes advantage of many more subsections of pictures this makes more work, but it is solved more easily.
v1.6 - Fully Documented
lids is a fully enabled API to create picture searches. Much like a TensorFlow, yet brought up from scratch without peeking, this package is a completely independent Neural Network.
The activity described by the code uses a single image to process against the dataset. If the dataset is populated with something similar or exactly the same, then it will bring back the tags associated with the image that were entered when the image was added to the population.
It will also return with what percent the image was predicted to be a match with.
This project was very fun. I was extremely excited once I got into it. To be able to curve yourself on the front lines of technology is something you shoul not refrain from if your talents can take you there. I promise.
There's nothing stopping you from being successful with this project in hand either if you plan on doing something with the web that needs to be lightweight and produce quickly, good results. Because this is simple to setup.
To add or search images just do this:
$object_var = new Tier();
$branch = new Branches();
$branch->origin = "/path/to/original/image/file.png";
$branch->keywords = array("first",$second,...);
$object_var->add_branch_img($branch);
that's it!
To relabel an image try this:
$branch = new Branches();
$branch->origin = "/path/to/original/image/file.png";
$node = $object_var->find_tier($branch);
$object_var->relabel_img($node, ["therein","we","go"]);
Remember to use the same Tier() object and you're all set! Done! Bravo!
File List:
Branches.php is a common list extensions acting as nodes
PNG.php is for creating the thumbnails
- resize_png() to scale and to change color depth
- ImageTrueColorToPalette2() to create image with new depth and scale
- crop_png() crops photos in 34 ways and saves to one file
- find_tier() resize image
Tier.php is for searching for the thumbnails
- insert_branch() used to add to common list of images
- add_branch_img() creates new thumbnail
- save_dataset() saves common list to file
- load_dataset() loads common list from file
- search_imgs() searches files for matches
- kernel_make() creates kernel sampling in 1x50 style
- label_search() get label of picture found
- relabel_img() find filename and relabel the image
Thanks for looking.