Annotation of /trunk/TODO

Parent Directory | Revision Log

---

Revision 1445 - (view) (download)

1 :	nseltzer	1445	THIS TODO HAS BEEN MOVED TO THE DIDEROT WIKI. PLEASE USE THAT ONE.
2 :
3 :	glk	1162	NOTE: GLK's approximate ranking of 8 most important tagged with
4 :	jhr	1115	[GLK:1], [GLK:2], ...
5 :
6 :	glk	1156	========================
7 :			SHORT TERM ============= (*needed* for streamlines & tractography)
8 :			========================
9 :	jhr	1115
10 :	glk	1338	[GLK:2] Add sequence types (needed for evals & evecs)
11 :	jhr	1115	syntax
12 :			types: ty '{' INT '}'
13 :			value construction: '{' e1 ',' … ',' en '}'
14 :			indexing: e '{' e '}'
15 :
16 :	glk	1338	[GLK:3] evals & evecs for symmetric tensor[2,2] and
17 :	glk	1162	tensor[3,3] (requires sequences)
18 :
19 :	glk	1156	ability to emit/track/record variables into dynamically re-sized
20 :	glk	1389	runtime output buffer
21 :	jhr	1115
22 :	glk	1442	[GLK:4] tensor fields from tensor images: Initially need at least
23 :			convolution on tensor[2,2] and tensor[3,3] (the same component-wise
24 :			convolution as for vectors).
25 :	jhr	1115
26 :	glk	1156	========================
27 :	glk	1162	SHORT-ISH TERM ========= (to make using Diderot less annoying to
28 :			======================== program in, and slow to execute)
29 :	jhr	1115
30 :	glk	1167	Allow ".ddro" file extensions in addition to ".diderot"
31 :
32 :	glk	1204	Be able to output values of type tensor[2,2] and tensor[3,3];
33 :			(currently only scalars & vectors). Want to add some regression tests
34 :			based on this and currently can't
35 :	glk	1167
36 :	glk	1338	[GLK:1] Proper handling of stabilize method
37 :	jhr	1115
38 :	glk	1442	Convolution on general tensor images (order > 2)
39 :
40 :	glk	1162	allow "*" to represent "modulate": per-component multiplication of
41 :			vectors, and vectors only (not tensors of order 2 or higher). Once
42 :			sequences are implemented this should be removed: the operation is not
43 :			invariant WRT basis so it is not a legit vector computation.
44 :
45 :			implicit type promotion of integers to reals where reals are
46 :			required (e.g. not exponentiation "^")
47 :
48 :	glk	1442	[Nick working on this] Save Diderot output to nrrd, instead of "mip.txt"
49 :	jhr	1115	For grid of strands, save to similarly-shaped array
50 :			For list of strands, save to long 1-D (or 2-D for non-scalar output) list
51 :			For ragged things (like tractography output), will need to save both
52 :			complete list of values, as well as list of start indices and lengths
53 :			to index into complete list
54 :
55 :	glk	1442	[GLK:6] ability to declare a field so that probe positions are
56 :	glk	1120	*always* "inside"; with various ways of mapping the known image values
57 :			to non-existant index locations. One possible syntax emphasizes that
58 :			there is a index mapping function that logically precedes convolution:
59 :	glk	1162	F = bspln3 ⊛ (img ◦ clamp)
60 :	glk	1120	F = bspln3 ⊛ (img ◦ repeat)
61 :			F = bspln3 ⊛ (img ◦ mirror)
62 :			where "◦" or "∘" is used to indicate function composition
63 :	jhr	1115
64 :	glk	1162	Level of differentiability in field type should be statement about how
65 :			much differentiation the program *needs*, rather than what the kernel
66 :			*provides*. The needed differentiability can be less than or equal to
67 :			the provided differentiability.
68 :
69 :	glk	1156	Use ∇⊗ etc. syntax
70 :			syntax [DONE]
71 :			typechecking
72 :			IL and codegen
73 :	jhr	1115
74 :	glk	1156	Add type aliases for color types
75 :			rgb = real{3}
76 :			rgba = real{4}
77 :	jhr	1115
78 :	glk	1389	Revisit how images are created within the language.
79 :			The "load" operator should probably go away, and its strangs
80 :			that strings are there only as a way to refer to nrrd filenames
81 :
82 :	jhr	1115	==============================
83 :	glk	1156	MEDIUM TERM ================== (*needed* for particles)
84 :	jhr	1115	==============================
85 :
86 :	glk	1442	[Lamont working on this] run-time birth of strands
87 :	jhr	1115
88 :			"initially" supports lists
89 :
90 :	glk	1254	"initially" supports lists of positions output from different
91 :			initalization Diderot program (or output from the same program;
92 :			e.g. using output of iso2d.diderot for one isovalue to seed the input
93 :			to another invocation of the same program)
94 :	jhr	1115
95 :	glk	1442	[Lamont working on this] Communication between strands: they have to
96 :			be able to learn each other's state (at the previous iteration).
97 :			Early version of this can have the network of neighbors be completely
98 :			static (for running one strand/pixel image computations). Later
99 :			version with strands moving through the domain will require some
100 :			spatial data structure to optimize discovery of neighbors.
101 :	jhr	1115
102 :	glk	1156	============================
103 :			MEDIUM-ISH TERM ============ (to make Diderot more useful/effective)
104 :			============================
105 :	jhr	1115
106 :	glk	1442	[GLK:5] Want code-generation working for tensors of order three.
107 :			Order three matters for edge detection in scalar fields (to get
108 :			second derivatives of gradient magnitude), second derivatives
109 :			of vector fields (for some feature extraction), and first
110 :			derivatives of diffusion tensor fields.
111 :
112 :	glk	1156	Python/ctypes interface to run-time
113 :	jhr	1115
114 :	glk	1156	support for Python interop and GUI
115 :	jhr	1115
116 :	glk	1162	Allow integer exponentiation ("^2") to apply to square matrices,
117 :			to represent repeated matrix multiplication
118 :
119 :	glk	1156	Put small 1-D and 2-D fields, when reconstructed specifically by tent
120 :			and when differentiation is not needed, into faster texture buffers.
121 :			test/illust-vr.diderot is good example of program that uses multiple
122 :			such 1-D fields basically as lookup-table-based function evaluation
123 :
124 :			extend norm (|exp|) to all tensor types [DONE for vectors and matrices]
125 :
126 :			determinant ("det") for tensor[3,3]
127 :
128 :	jhr	1115	add ":" for tensor dot product (contracts out two indices
129 :			instead of one like •), valid for all pairs of tensors with
130 :			at least two indices
131 :
132 :	glk	1156	test/uninit.diderot:
133 :			documents need for better compiler error messages when output variables
134 :			are not initialized; the current messages are very cryptic
135 :	jhr	1115
136 :			want: warnings when "D" (reserved for differentiation) is declared as
137 :			a variable name (get confusing error messages now)
138 :
139 :	glk	1156	==============================
140 :			LONG TERM ==================== (make Diderot more interesting/attractive from
141 :			============================== a research standpoint)
142 :	jhr	1115
143 :	glk	1156	IL support for higher-order tensor values (matrices, etc).
144 :			tensor construction [DONE]
145 :			tensor indexing [DONE]
146 :			tensor slicing
147 :	jhr	1115
148 :	glk	1156	Better handling of variables that determines the scope of a variable
149 :			based on its actual use, instead of where the user defined it. So,
150 :			for example, we should lift strand-invariant variables to global
151 :			scope. Also prune out useless variables, which should include field
152 :			variables after the translation to mid-il.
153 :
154 :			test/vr-kcomp2.diderot: Add support for code like
155 :			(F1 if x else F2)@pos
156 :			This will require duplication of the continuation of the conditional
157 :			(but we should only duplicate over the live-range of the result of the
158 :			conditional.
159 :
160 :	glk	1338	[GLK:7] Want: non-trivial field expressions & functions.
161 :	glk	1162	scalar fields from scalar fields F and G:
162 :			field#0(2)[] X = (sin(F) + 1.0)/2;
163 :			field#0(2)[] X = F*G;
164 :			scalar field of vector field magnitude:
165 :	glk	1156	image(2)[2] Vimg = load(...);
166 :			field#0(2)[] Vlen = |Vimg ⊛ bspln3|;
167 :	glk	1162	field of normalized vectors (for LIC and vector field feature extraction)
168 :			field#2(2)[2] F = ...
169 :			field#0(2)[2] V = normalize(F);
170 :			scalar field of gradient magnitude (for edge detection))
171 :	glk	1156	field#2(2)[] F = Fimg ⊛ bspln3;
172 :			field#0(2)[] Gmag = |∇F|;
173 :	glk	1162	scalar field of squared gradient magnitude (simpler to differentiate):
174 :	glk	1156	field#2(2)[] F = Fimg ⊛ bspln3;
175 :			field#0(2)[] Gmsq = ∇F•∇F;
176 :	glk	1162	There is value in having these, even if the differentiation of them is
177 :			not supported (hence the indication of "field#0" for these above)
178 :	glk	1156
179 :	jhr	1195	Introduce region types (syntax region(d), where d is the dimension of the
180 :			region. One useful operator would be
181 :			dom : field#k(d)[s] -> region(d)
182 :			Then the inside test could be written as
183 :			pos ∈ dom(F)
184 :			We could further extend this approach to allow geometric definitions of
185 :			regions. It might also be useful to do inside tests in world space,
186 :			instead of image space.
187 :
188 :	glk	1162	co- vs contra- index distinction
189 :	glk	1156
190 :	glk	1162	Permit field composition:
191 :	glk	1156	field#2(3)[3] warp = bspln3 ⊛ warpData;
192 :			field#2(3)[] F = bspln3 ⊛ img;
193 :			field#2(3)[] Fwarp = F ◦ warp;
194 :	glk	1162	So Fwarp(x) = F(warp(X)). Chain rule can be used for differentation.
195 :			This will be instrumental for expressing non-rigid registration
196 :			methods (but those will require co-vs-contra index distinction)
197 :	glk	1156
198 :	glk	1155	Allow the convolution to be specified either as a single 1D kernel
199 :			(as we have it now):
200 :			field#2(3)[] F = bspln3 ⊛ img;
201 :			or, as a tensor product of kernels, one for each axis, e.g.
202 :			field#0(3)[] F = (bspln3 ⊗ bspln3 ⊗ tent) ⊛ img;
203 :	glk	1212	This is especially important for things like time-varying fields
204 :			and the use of scale-space in field visualization: one axis of the
205 :			must be convolved with a different kernel during probing.
206 :			What is very unclear is how, in such cases, we should notate the
207 :			gradient, when we only want to differentiate with respect to some
208 :			subset of the axes. One ambitious idea would be:
209 :	glk	1162	field#0(3)[] Ft = (bspln3 ⊗ bspln3 ⊗ tent) ⊛ img; // 2D time-varying field
210 :	glk	1204	field#0(2)[] F = lambda([x,y], Ft([x,y,42.0])) // restriction to time=42.0
211 :			vec2 grad = ∇F([x,y]); // 2D gradient
212 :	glk	1155
213 :	glk	1162	representation of tensor symmetry
214 :	jhr	1115	(have to identify the group of index permutations that are symmetries)
215 :
216 :			dot works on all tensors
217 :
218 :			outer works on all tensors
219 :
220 :	glk	1204	Help for debugging Diderot programs: need to be able to uniquely
221 :			identify strands, and for particular strands that are known to behave
222 :			badly, do something like printf or other logging of their computations
223 :			and updates.
224 :
225 :			Permit writing dimensionally general code: Have some statement of the
226 :			dimension of the world "W" (or have it be learned from one particular
227 :			field of interest), and then able to write "vec" instead of
228 :			"vec2/vec3", and perhaps "tensor[W,W]" instead of
229 :			"tensor[2,2]/tensor[3,3]"
230 :
231 :			Traits: all things things that have boilerplate code (especially
232 :			volume rendering) should be expressed in terms of the unique
233 :			computational core. Different kinds of streamline/tractography
234 :			computation will be another example, as well as particle systems.
235 :
236 :	jhr	1115	Einstein summation notation
237 :
238 :			"tensor comprehension" (like list comprehension)
239 :
240 :	glk	1204	Fields coming from different sources of data:
241 :			* triangular or tetrahedral meshes over 2D or 3D domains (of the
242 :			source produced by finite-element codes; these will come with their
243 :			own specialized kinds of reconstruction kernels, called "basis
244 :			functions" in this context)
245 :			* Large point clouds, with some radial basis function around each point,
246 :			which will be tuned by parameters of the point (at least one parameter
247 :			giving some notion of radius)
248 :
249 :	jhr	1115	======================
250 :			BUGS =================
251 :			======================
252 :
253 :			test/zslice2.diderot:
254 :			// HEY (bug) bspln5 leads to problems ...
255 :			// uncaught exception Size [size]
256 :			// raised at c-target/c-target.sml:47.15-47.19
257 :			//field#4(3)[] F = img ⊛ bspln5;
258 :	glk	1336

---

root@smlnj-gforge.cs.uchicago.edu	ViewVC Help
Powered by ViewVC 1.0.0