According to the form of the data set provided, you need to take 10 image samples of digital gestures/hand-Signs under 10 different backgrounds（100 image samples totally） and save in the folders named A0 to A9. These new data will be collected before the last class. You should design and build your own gesture image recognition network and complete the training test using the provided data set. In the last class, images will be collected for the final performance test, and scores will be given according to the test results. In principle, it is required that the recognition rate should be higher than 50%. Submission content:
- The PPT report must be completed in English, and the main content (besides cover, catalogue, thanks and section title pages) shall not be less than 20 pages, which is used for experience exchange in class. (completed before class and can be modified and supplemented after class)
- For the text scheme report, you can choose your own language, but the PPT report must be completed in English, no less than 5000 words. See the report requirements for details (collect process materials(training curves, data, intermediate result .etc, try using “Print Screen”) before class and complete after class)
- The source code, you should mark the parts you have completed, add detailed comments, and only keep py，Ipynb and other text form files. The size shall not exceed 5MB. (Your program can run completely before class including train and test. You will be checked one by one. You should pack and submit source code after class) After class submission method: text report, PPT report, source code, packaged in a compressed package. Name: student number_ Name_ FinalProject. rar(.zip)。 Send to the specified mailbox (the same for hand in homework).
Task requirements: Training data status
-
Data segmentation strategy(Train, validation, test).
-
Task process (including details like: hyper parameter validation, training curves, result analysis, structure, regularization, parameter tuning process and problem analysis and design thoughts) 2.1 network structure design and construction 2.2 selection and validation of hyper parameters 2.3 the first network training and test 2.4 do some adjustment 2.5 the second network training and test
-
Classroom test (complete after class before final submission) Testing process, Test performance, whether it meets the expectation and cause analysis.
-
Course thoughts