problem about training the SOS
Opened this issue · 0 comments
Cocofeat commented
when I train the SOS, I found the dice results of small organs are always equal to zero.
if results['heatmap'][0, j, :, :, :].max() > 0.6:
z, x, y = results['location'][j]
smallOutput[j, HSi+z-4:HSi+z+4, x-32:x+32, y-32:y+32] += F.sigmoid(results['small_result'][0, j, :, :, :])
small_counter[j, HS*i+z-4:HS*i+z+4, x-32:x+32, y-32:y+32] += small_one_count
I found that the condition of the "if results['heatmap'][0, j, :, :, :].max() > 0.6" always failed. Can you give some suggestions? Thanks!