1. Frame-by-frame analyzer:

• Find characters and their animation states
• Find frame interval analyzed (
• Write to .csv (maybe local array and write off at the end?)

2. UI, youtube link -> video -> stats -> database

• What do we need from user?

•   Names of players, date, tournament (round?)
  

• How do we validate this data before putting it in the database?

3. Derived tables

• Primary keys: players, date, tournament/round, game #, frame interval
• Generated by engine: animation states
• From this, derive tables:
• Lowest level: button inputs (-> A, etc)
• Higher: move names (fsmash, etc)
• Highest: above w/ tech (wavedash, shffl, etc)

4. Low-level reports:

• combos
• situational decision making

5. Higher-level reports

• Stage control
• Edgeguarding efficiency
• Decision trees/Option coverage/Reads
• Spacing

6. Metagame reports

• What works against certain characters?
• Matchup records
• Stage data

Known limitations:

• YouTube: <30 fps
• DI can't be identified (right?)
• Fast inputs (worst case: 7 game frames btw youtube frames)

Progress:

• Working on basic object detection, identifying pikachu and fox in game starting positions