mirror-github-bspeice/cuburn
1
0
Fork 0
mirror of https://github.com/stevenrobertson/cuburn.git synced 2025-11-03 18:00:55 -05:00
Code Issues Actions Packages Projects Releases Wiki Activity

Make DE better

Browse Source
This commit is contained in:
Steven Robertson
2011-12-10 16:24:49 -05:00
parent c59829ad86
commit 12655b8611
3 changed files with 79 additions and 101 deletions
Download Patch File Download Diff File Expand all files Collapse all files

97
cuburn/code/filtering.py
View File

@ -85,9 +85,9 @@ void colorclip(float4 *pixbuf, float gamma, float vibrancy, float highpow,
} }
#define W 21 // Filter width (regardless of standard deviation chosen) #define W 15 // Filter width (regardless of standard deviation chosen)
#define W2 10 // Half of filter width, rounded down #define W2 7 // Half of filter width, rounded down
#define FW 52 // Width of local result storage (NW+W2+W2) #define FW 46 // Width of local result storage (NW+W2+W2)
#define FW2 (FW*FW) #define FW2 (FW*FW)
__shared__ float de_r[FW2], de_g[FW2], de_b[FW2], de_a[FW2]; __shared__ float de_r[FW2], de_g[FW2], de_b[FW2], de_a[FW2];
@ -116,12 +116,12 @@ void logscale(float4 *pixbuf, float4 *outbuf, float k1, float k2) {
// See helpers/filt_err.py for source of these values. // See helpers/filt_err.py for source of these values.
#define MIN_SD 0.23299530f #define MAX_SCALE -0.12f
#define MAX_SD 4.33333333f #define MIN_SCALE -9.2103404f
__global__ __global__
void density_est(float4 *pixbuf, float4 *outbuf, void density_est(float4 *pixbuf, float4 *outbuf,
float est_sd, float neg_est_curve, float est_min, float scale_coeff, float est_curve, float edge_clamp,
float k1, float k2, int height, int stride) { float k1, float k2, int height, int stride) {
for (int i = threadIdx.x + 32*threadIdx.y; i < FW2; i += 32) for (int i = threadIdx.x + 32*threadIdx.y; i < FW2; i += 32)
de_r[i] = de_g[i] = de_b[i] = de_a[i] = 0.0f; de_r[i] = de_g[i] = de_b[i] = de_a[i] = 0.0f;
@ -164,57 +164,46 @@ void density_est(float4 *pixbuf, float4 *outbuf,
// Base index of destination for writes // Base index of destination for writes
int si = (threadIdx.y + W2) * FW + threadIdx.x + W2; int si = (threadIdx.y + W2) * FW + threadIdx.x + W2;
// Calculate standard deviation of Gaussian kernel. The base SD is // Calculate scaling coefficient for the Gaussian kernel. This
// then scaled in inverse proportion to the density of the point // does not match with a normal Gaussian; it just fits with
// being scaled. // flam3's implementation.
float sd = est_sd * powf(den+1.0f, neg_est_curve); float scale = powf(den, est_curve) * scale_coeff;
// And for the gradient...
float diag_sd = est_sd * powf(diag_mag+1.0f, neg_est_curve);
// If the gradient SD is smaller than the minimum SD, we're probably // If the gradient scale is smaller than the minimum scale, we're
// on a strong edge; blur with a standard deviation around 1px. // probably on a strong edge; blur slightly.
if (diag_sd < MIN_SD && diag_sd < sd) { if (diag_mag > den * 2.0f) {
sd = 0.3333333f; scale = max(-9.0f, scale);
// Uncomment to see which pixels are being clamped // Uncomment to see which pixels are being clamped
// de_g[si] = 1.0f; // de_g[si] = 1.0f;
} }
// Clamp the final standard deviation.
sd = fminf(MAX_SD, fmaxf(sd, est_min));
// Below a certain threshold, only one coeffecient would be // Below a certain threshold, only one coeffecient would be
// retained anyway; we hop right to it. // retained anyway; we hop right to it.
if (sd <= MIN_SD) { if (scale <= MIN_SCALE) {
de_add(si, 0, 0, in); de_add(si, 0, 0, in);
} else { } else {
// These polynomials approximates the sum of the filters // These polynomials approximates the reciprocal of the sum of
// with the clamping logic used here. See helpers/filt_err.py. // all retained filter coefficients. See helpers/filt_err.py.
float filtsum; float filtsum;
if (sd < 0.75f) { if (scale < -1.1f) {
filtsum = -352.25061035f; filtsum = 5.20066078e-06f;
filtsum = filtsum * sd + 1117.09680176f; filtsum = filtsum * scale + 2.14025771e-04f;
filtsum = filtsum * sd + -1372.48864746f; filtsum = filtsum * scale + 3.62761668e-03f;
filtsum = filtsum * sd + 779.15478516f; filtsum = filtsum * scale + 3.21970172e-02f;
filtsum = filtsum * sd + -164.04229736f; filtsum = filtsum * scale + 1.54297248e-01f;
filtsum = filtsum * sd + -12.04892635f; filtsum = filtsum * scale + 3.42210710e-01f;
filtsum = filtsum * sd + 9.04126644f; filtsum = filtsum * scale + 3.06015890e-02f;
filtsum = filtsum * sd + 0.10304667f; filtsum = filtsum * scale + 1.33724615e-01f;
} else { } else {
filtsum = 0.01162011f; filtsum = -1.23516649e-01f;
filtsum = filtsum * sd + -0.21552004f; filtsum = filtsum * scale + -5.14862895e-01f;
filtsum = filtsum * sd + 1.66545594f; filtsum = filtsum * scale + -8.61198902e-01f;
filtsum = filtsum * sd + -7.00809765f; filtsum = filtsum * scale + -7.41916001e-01f;
filtsum = filtsum * sd + 17.55487633f; filtsum = filtsum * scale + -3.51667106e-01f;
filtsum = filtsum * sd + -26.80626106f; filtsum = filtsum * scale + -9.07439440e-02f;
filtsum = filtsum * sd + 30.61903954f; filtsum = filtsum * scale + -3.30008656e-01f;
filtsum = filtsum * sd + -12.00870514f; filtsum = filtsum * scale + -4.78249392e-04f;
filtsum = filtsum * sd + 2.46708894f;
} }
float filtscale = 1.0f / filtsum;
// The reciprocal SD scaling coeffecient in the Gaussian
// exponent: exp(-x^2/(2*sd^2)) = exp2f(x^2*rsd)
float rsd = -0.5f * CUDART_L2E_F / (sd * sd);
for (int jj = 0; jj <= W2; jj++) { for (int jj = 0; jj <= W2; jj++) {
float jj2f = jj; float jj2f = jj;
@ -222,9 +211,8 @@ void density_est(float4 *pixbuf, float4 *outbuf,
float iif = 0; float iif = 0;
for (int ii = 0; ii <= jj; ii++) { for (int ii = 0; ii <= jj; ii++) {
float coeff = expf((jj2f + iif * iif) * scale)
float coeff = exp2f((jj2f + iif * iif) * rsd) * filtsum;
* filtscale;
if (coeff < 0.0001f) break; if (coeff < 0.0001f) break;
iif += 1; iif += 1;
@ -317,15 +305,12 @@ class Filtering(object):
t = fun(dsrc, ddst, k1, k2, t = fun(dsrc, ddst, k1, k2,
block=(512, 1, 1), grid=(nbins/512, 1), stream=stream) block=(512, 1, 1), grid=(nbins/512, 1), stream=stream)
else: else:
# flam3_gaussian_filter() uses an implicit standard deviation of scale_coeff = np.float32(-(1 + cp.de.radius(t)) ** -2.0)
# 0.5, but the DE filters scale filter distance by the default est_curve = np.float32(2 * cp.de.curve(t))
# spatial support factor of 1.5, so the effective base SD is # TODO: experiment with this
# (0.5/1.5)=1/3. edge_clamp = np.float32(2.0)
est_sd = np.float32(cp.de.radius(t) / 3.)
neg_est_curve = np.float32(-cp.de.curve(t))
est_min = np.float32(cp.de.minimum(t) / 3.)
fun = self.mod.get_function("density_est") fun = self.mod.get_function("density_est")
fun(dsrc, ddst, est_sd, neg_est_curve, est_min, k1, k2, fun(dsrc, ddst, scale_coeff, est_curve, edge_clamp, k1, k2,
np.int32(info.acc_height), np.int32(info.acc_stride), np.int32(info.acc_height), np.int32(info.acc_stride),
block=(32, 32, 1), grid=(info.acc_width/32, 1), stream=stream) block=(32, 32, 1), grid=(info.acc_width/32, 1), stream=stream)

2
cuburn/genome.py
View File

@ -116,7 +116,7 @@ class RenderInfo(object):
# Maximum width of DE and other spatial filters, and thus in turn the # Maximum width of DE and other spatial filters, and thus in turn the
# amount of padding applied. Note that, for now, this must not be changed! # amount of padding applied. Note that, for now, this must not be changed!
# The filtering code makes deep assumptions about this value. # The filtering code makes deep assumptions about this value.
gutter = 22 gutter = 15
# TODO: for now, we always throw away the alpha channel before writing. # TODO: for now, we always throw away the alpha channel before writing.
# All code is in place to not do this, we just need to find a way to expose # All code is in place to not do this, we just need to find a way to expose

81
helpers/filt_err.py
View File

@ -1,7 +1,7 @@
import numpy as np import numpy as np
# The maximum number of coeffecients that will ever be retained on the device # The maximum number of coeffecients that will ever be retained on the device
FWIDTH = 21 FWIDTH = 15
# The number of points on either side of the center in one dimension # The number of points on either side of the center in one dimension
F2 = int(FWIDTH/2) F2 = int(FWIDTH/2)
@ -12,92 +12,85 @@ COEFF_EPS = 0.0001
dists2d = np.fromfunction(lambda i, j: np.hypot(i-F2, j-F2), (FWIDTH, FWIDTH)) dists2d = np.fromfunction(lambda i, j: np.hypot(i-F2, j-F2), (FWIDTH, FWIDTH))
dists = dists2d.flatten() dists = dists2d.flatten()
# A flam3 estimator radius corresponds to a Gaussian filter with a standard
# deviation of 1/3 the radius. We choose 13 as an arbitrary upper bound for the
# max filter radius. The filter should reject larger radii.
MAX_SD = 13 / 3.
# The minimum estimator radius can be set as low as 0, but below a certain # This translates to a cap on DE filter radius of 50. Even this fits very
# radius only one coeffecient is retained. Since things get unstable near 0, # comfortably within the chosen COEFF_EPS.
# we explicitly set a minimum threshold below which no coeffecients are MAX_SCALE = -3/25.
# retained.
MIN_SD = np.sqrt(-1 / (2 * np.log(COEFF_EPS)))
# Using two predicated three-term approximations is much more accurate than # When the scale is above this value, we'd be directly clamping to one bin
# using a very large number of terms, due to nonlinear behavior at low SD. MIN_SCALE = np.log(0.0001)
# Everything above this SD uses one approximation; below, another.
SPLIT_SD = 0.75
# The lower endpoints are undershot by this proportion to reduce error # Everything above this scale uses one approximation; below, another.
UNDERSHOOT = 0.98 SPLIT_SCALE = -1.1
sds_hi = np.linspace(SPLIT_SD * UNDERSHOOT, MAX_SD, num=1000) # The upper endpoints are overshot by this proportion to reduce error
sds_lo = np.linspace(MIN_SD * UNDERSHOOT, SPLIT_SD, num=1000) OVERSHOOT = 1.01
print 'At MIN_SD = %g, these are the coeffs:' % MIN_SD # No longer 'scale'-related, but we call it that anyway
print np.exp(dists2d**2 / (-2 * MIN_SD ** 2)) scales_hi = np.linspace(SPLIT_SCALE, MAX_SCALE * OVERSHOOT, num=1000)
scales_lo = np.linspace(MIN_SCALE, SPLIT_SCALE * OVERSHOOT, num=1000)
def eval_sds(sds, name, nterms): def eval_scales(scales, name, nterms):
# Calculate the filter sums at each coordinate # Calculate the filter sums at each coordinate
sums = [] sums = []
for sd in sds: for scale in scales:
coeffs = np.exp(dists**2 / (-2 * sd ** 2)) coeffs = np.exp(dists**2 * scale)
# Note that this sum is the sum of all coordinates, though it should # Note that this sum is the sum of all coordinates, though it should
# actually be the result of the polynomial approximation. We could do # actually be the result of the polynomial approximation. We could do
# a feedback loop to improve accuracy, but I don't think the difference # a feedback loop to improve accuracy, but I don't think the difference
# is worth worrying about. # is worth worrying about.
sum = np.sum(coeffs) sum = np.sum(coeffs)
sums.append(np.sum(filter(lambda v: v / sum > COEFF_EPS, coeffs))) sums.append(1./np.sum(filter(lambda v: v / sum > COEFF_EPS, coeffs)))
print 'Evaluating %s:' % name print 'Evaluating %s:' % name
poly, resid, rank, sing, rcond = np.polyfit(sds, sums, nterms, full=True) poly, resid, rank, sing, rcond = np.polyfit(scales, sums, nterms, full=True)
print 'Fit for %s:' % name, poly, resid, rank, sing, rcond print 'Fit for %s:' % name, poly, resid, rank, sing, rcond
return sums, poly return sums, poly
import matplotlib.pyplot as plt import matplotlib.pyplot as plt
sums_hi, poly_hi = eval_sds(sds_hi, 'hi', 8) sums_hi, poly_hi = eval_scales(scales_hi, 'hi', 7)
sums_lo, poly_lo = eval_sds(sds_lo, 'lo', 7) sums_lo, poly_lo = eval_scales(scales_lo, 'lo', 7)
num_undershoots = len(filter(lambda v: v < SPLIT_SD, sds_hi)) num_overshoots = len(filter(lambda v: v > MAX_SCALE, scales_hi))
sds_hi = sds_hi[num_undershoots:] scales_hi = scales_hi[num_overshoots:]
sums_hi = sums_hi[num_undershoots:] sums_hi = sums_hi[num_overshoots:]
num_undershoots = len(filter(lambda v: v < MIN_SD, sds_lo)) num_overshoots = len(filter(lambda v: v > SPLIT_SCALE, scales_lo))
sds_lo = sds_lo[num_undershoots:] scales_lo = scales_lo[num_overshoots:]
sums_lo = sums_lo[num_undershoots:] sums_lo = sums_lo[num_overshoots:]
polyf_hi = np.float32(poly_hi) polyf_hi = np.float32(poly_hi)
vals_hi = np.polyval(polyf_hi, sds_hi) vals_hi = np.polyval(polyf_hi, scales_hi)
polyf_lo = np.float32(poly_lo) polyf_lo = np.float32(poly_lo)
vals_lo = np.polyval(polyf_lo, sds_lo) vals_lo = np.polyval(polyf_lo, scales_lo)
def print_filt(filts): def print_filt(filts):
print ' filtsum = %4.8ff;' % filts[0] print ' filtsum = %4.8ef;' % filts[0]
for f in filts[1:]: for f in filts[1:]:
print ' filtsum = filtsum * sd + % 16.8ff;' % f print ' filtsum = filtsum * scale + % 16.8ef;' % f
print '\n\nFor your convenience:' print '\n\nFor your convenience:'
print '#define MIN_SD %.8f' % MIN_SD print '#define MIN_SCALE %.8gf' % MIN_SCALE
print '#define MAX_SD %.8f' % MAX_SD print '#define MAX_SCALE %.8gf' % MAX_SCALE
print 'if (sd < %g) {' % SPLIT_SD print 'if (scale < %gf) {' % SPLIT_SCALE
print_filt(polyf_lo) print_filt(polyf_lo)
print '} else {' print '} else {'
print_filt(polyf_hi) print_filt(polyf_hi)
print '}' print '}'
sds = np.concatenate([sds_lo, sds_hi]) scales = np.concatenate([scales_lo, scales_hi])
sums = np.concatenate([sums_lo, sums_hi]) sums = np.concatenate([sums_lo, sums_hi])
vals = np.concatenate([vals_lo, vals_hi]) vals = np.concatenate([vals_lo, vals_hi])
fig = plt.figure() fig = plt.figure()
ax = fig.add_subplot(1,1,1) ax = fig.add_subplot(1,1,1)
ax.plot(sds, sums) ax.plot(scales, sums)
ax.plot(sds, vals) ax.plot(scales, vals)
ax.set_xlabel('stdev') ax.set_xlabel('stdev')
ax.set_ylabel('filter sum') ax.set_ylabel('filter sum')
ax = ax.twinx() ax = ax.twinx()
ax.plot(sds, [abs((s-v)/v) for s, v in zip(sums, vals)]) ax.plot(scales, [abs((s-v)/v) for s, v in zip(sums, vals)])
ax.set_ylabel('rel err') ax.set_ylabel('rel err')
plt.show() plt.show()