-
Frame Selection (Part 1)
- get_frames : function to get frames from video
- filter_frames : function to filter out bad/blurry frames
-
Product Detection (Part 2)
-
Training
- download/gather appropriate dataset
- visualize the dataset
- choose a model
- training the model
- evaluate the model
- save the model
-
Inference
- model creation code
- model weights loading code
- model's output to xmin,ymin,xmax,ymax bounding box coordinates for each detection
- bounding box to crops array
-
Evaluation
- load the model using the functions in inference folder
- evaluate the model's performance on some dataset and metric
-
-
Crop Selection (Part 3)
- filter_crops : function to filter out bad/unusable/blurry/very low res crops
- very low res crops
- blurry crops
-
Product Recognition (Part 4)
-
Training
- download rp2k dataset
- loading and clean the dataset
- visualize the dataset
- choose a model
- train the model
- evaluate the model (on some appropriate dataset and metric)
- save the model (only the backbone feature extractor)
-
Inference
- model creation code
- model weights loading code
- get_embedding : function for getting feature vectors
- get_simmilarity_scores : function which takes two feature vectors and compare then and gives the simmilarity scores 0 to 1
- get_knn : returns sklearn K nearest neighbours model
-
Evaluation
- load the model using the functions in inference folder
- evaluate the model's performance on some dataset and metric
-
-
Pipeline (Part 5)
- create_database : function to create database of products
- load_database : function to load products database
- image_to_products : find all product skus in a image and return crops with their guessed sku id
- video_to_products : find all product skus in a video and return crops with their guessed sku id