3 * generate proper "move" steps in SGF when possible
4 * use the grid region only for quantization (autodetecting best board
5 color from that region)
6 * calculate default sample_radius from board size on image
7 + compensation of inacuracies:
8 * provide diagnostic about recognition errors based on expectation
10 * try to filter out erroneous add/clear when they are undone later
11 (keep a list of positions known to vary, and confidence stats)
12 * add SGF comments about (non-)confidence of decoded moves when needed
15 * make the solution able to handle 13x13 and 9x9 without changing the code
16 (esp. find gocam settings for 9x9 photos, it seems just confused)
17 * handling particular boards like hoshi-delimited 13x13 on a 19x19 board
18 * provide a command-line tool to get the job done without the need to
20 * maybe provide a gui for interactive tuning
23 * port the Image2SGF's stone-detection algo into gocam to avoid dep on perl
24 and hopefully get better perf