update readme.md

README.md

@@ -4,17 +4,14 @@
 <br></br>
 
 ## Application Overview
-When using the provided network in places like cafe, people have to search for the matching WIFI network ID in the available WIFI list and type the PW manually.
-**This Android application will enable user to connect to WIFI automatically by just simply scanning the picture that contains WIFI ID and PW**
-
-<img src="./image_readme/wifi.jpg" width="450px"></img>
+When using the provided network in places like cafe, people have to search for the matching wifi SSID in the available Wifi list and type the password manually. This Android application will enable user to connect to wifi automatically by just simply scanning the picture that contains wifi SSID and password.
 
 <br></br>
 
 ## Process Pipeline
-1. [Input] Picture that contains WIFI ID and PW. The picture can contain letters other than the ID and PW, and the letters can be hand-written.
-2. Detect the words and their positions using OCR pretrained model. 
-3. From the detected information in step 2, extract WIFI ID and PW.
+1. Input: picture that contains wifi SSID and password taken by user.
+2. Text detection (description string, bounding boxes) using Google AI Cloud Vision API. 
+3. Output: extracted wifi SSID and password from the detection result, which is required to connect to wifi.
 4. Using Android network API, connect to WIFI automatically using retrieved ID and PW.
 
 <br></br>
@@ -23,24 +20,5 @@ When using the provided network in places like cafe, people have to search for t
 <img src="./image_readme/Use Flow.png"></img>
 
 <br></br>
-
-## Considerations for Implementation
-### Evaluation
-Though we use pretrained model for OCR, we have to post process the ID and PW information from the OCR model output. We will evaluate out application by measuring the success rate of "Extracting both correct ID and PW from arbitrary WIFI information picture".
-
-### Input data
-Though we use pretrained model, we need input data for making a good post processing strategy and evaluation.
-We will first collect some real WIFI information picture data by Hand collecting and crawl from Google image search.
-Since the images we require are very specific, the amount of data that can be collected by above ways might not be large. However, the reason we need input data is not for training but for evaluation and deciding post processing strategy. If the collected data is too small even for above objectives, we will search for other ways to enlarge the data. (e.g. generation based on collected real data)
-
-### OCR model
-We will modularize here, and first use OCR cloud API. 
-This will require network connection, so we will try applying an on-device model if time allows. 
-Model output will be collection of 
-{Detected characters, Bounding box position and size}s 
-
-### Post processing for ID and PW extraction
-We will first try deterministic way for this, using approaches listed below. If it is too unsuccessful, we should search for other approaches.
-- Word position and size. (Centered, Big, SSID is usually followed by Password, etc)
-- Field tag. (ID, PW, etc)
-- Match with retrieved available Wi-fi SSIDs.
+## Project document:
+https://docs.google.com/document/d/1rUMfPKWnKN4waLV1V4iEL9Xn7siuPR_zNOVPaEmiBW4