Detecting Vairous products in A Retail Shop

Challenges

• Thousands of products and hundreds of vaients of different product
• Cannot label different products individually
• Text is found in various fonts, orientation and sizes
• Each frame of video has to be processed

Strategy

• We to use written content on product
• We also use object classification of different images in imagenet
• It would be possible to map the classification and text to product in a catalogue(for a real case scenario)

Solutions

• Using imagenet trained model
• Using CRAFT to read labels and texts on each products

Results

Replication

• Use requirement.txt to install dependencies
• Also download 'craft_mlt_25k.pth'
• Read about CRAFT algorithm if problem arises

Limitations

• Best suited for single photo, as the CRAFT method is slow in CPU

Future Work

• To create a faster working algorithm than CRAFT