+
+
+
+
+
+
+ target
+ 160×128 px · 61,440 values
+
+ ∞ dB
+
+
+
+
+
+ leaderboard · PSNR dB, higher is better · steps · speed
PSNR (dB) over time
$ tail -f train.log
+
+
+ how it got there · and, above, where each network is still wrong (brighter = bigger error)
+ This is a race between four tiny neural networks with the same theoretical capacity learning the same painting with the same compute. + Each has exactly 8,707 trainable numbers and the same CPU time. +
+ +