Mercurial > cortex

annotate org/vision.org @ 430:5205535237fb

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
fix skew in self-organizing-touch, work on thesis.
author Robert McIntyre <rlm@mit.edu>
date Sat, 22 Mar 2014 16:10:34 -0400
parents 5405f369f4a0
children 3401053124b0
rev   line source
rlm@34 1 #+title: Simulated Sense of Sight
rlm@23 2 #+author: Robert McIntyre
rlm@23 3 #+email: rlm@mit.edu
rlm@38 4 #+description: Simulated sight for AI research using JMonkeyEngine3 and clojure
rlm@34 5 #+keywords: computer vision, jMonkeyEngine3, clojure
rlm@23 6 #+SETUPFILE: ../../aurellem/org/setup.org
rlm@23 7 #+INCLUDE: ../../aurellem/org/level-0.org
rlm@23 8 #+babel: :mkdirp yes :noweb yes :exports both
rlm@23 9
ocsenave@264 10 * JMonkeyEngine natively supports multiple views of the same world.
ocsenave@264 11
rlm@212 12 Vision is one of the most important senses for humans, so I need to
rlm@212 13 build a simulated sense of vision for my AI. I will do this with
rlm@306 14 simulated eyes. Each eye can be independently moved and should see its
rlm@212 15 own version of the world depending on where it is.
rlm@212 16
rlm@306 17 Making these simulated eyes a reality is simple because jMonkeyEngine
rlm@306 18 already contains extensive support for multiple views of the same 3D
rlm@218 19 simulated world. The reason jMonkeyEngine has this support is because
rlm@218 20 the support is necessary to create games with split-screen
rlm@218 21 views. Multiple views are also used to create efficient
rlm@212 22 pseudo-reflections by rendering the scene from a certain perspective
rlm@212 23 and then projecting it back onto a surface in the 3D world.
rlm@212 24
rlm@218 25 #+caption: jMonkeyEngine supports multiple views to enable split-screen games, like GoldenEye, which was one of the first games to use split-screen views.
rlm@212 26 [[../images/goldeneye-4-player.png]]
rlm@212 27
ocsenave@264 28 ** =ViewPorts=, =SceneProcessors=, and the =RenderManager=.
rlm@306 29 # =ViewPorts= are cameras; =RenderManger= takes snapshots each frame.
ocsenave@264 30 #* A Brief Description of jMonkeyEngine's Rendering Pipeline
rlm@212 31
rlm@213 32 jMonkeyEngine allows you to create a =ViewPort=, which represents a
rlm@213 33 view of the simulated world. You can create as many of these as you
rlm@213 34 want. Every frame, the =RenderManager= iterates through each
rlm@213 35 =ViewPort=, rendering the scene in the GPU. For each =ViewPort= there
rlm@213 36 is a =FrameBuffer= which represents the rendered image in the GPU.
rlm@151 37
rlm@306 38 #+caption: =ViewPorts= are cameras in the world. During each frame, the =RenderManager= records a snapshot of what each view is currently seeing; these snapshots are =FrameBuffer= objects.
ocsenave@265 39 #+ATTR_HTML: width="400"
ocsenave@272 40 [[../images/diagram_rendermanager2.png]]
ocsenave@262 41
rlm@213 42 Each =ViewPort= can have any number of attached =SceneProcessor=
rlm@213 43 objects, which are called every time a new frame is rendered. A
rlm@306 44 =SceneProcessor= receives its =ViewPort's= =FrameBuffer= and can do
rlm@219 45 whatever it wants to the data. Often this consists of invoking GPU
rlm@219 46 specific operations on the rendered image. The =SceneProcessor= can
rlm@219 47 also copy the GPU image data to RAM and process it with the CPU.
rlm@151 48
ocsenave@264 49 ** From Views to Vision
ocsenave@264 50 # Appropriating Views for Vision.
rlm@151 51
ocsenave@264 52 Each eye in the simulated creature needs its own =ViewPort= so that
rlm@213 53 it can see the world from its own perspective. To this =ViewPort=, I
rlm@306 54 add a =SceneProcessor= that feeds the visual data to any arbitrary
rlm@213 55 continuation function for further processing. That continuation
rlm@213 56 function may perform both CPU and GPU operations on the data. To make
rlm@213 57 this easy for the continuation function, the =SceneProcessor=
rlm@306 58 maintains appropriately sized buffers in RAM to hold the data. It does
rlm@218 59 not do any copying from the GPU to the CPU itself because it is a slow
rlm@218 60 operation.
rlm@214 61
rlm@213 62 #+name: pipeline-1
rlm@213 63 #+begin_src clojure
rlm@113 64 (defn vision-pipeline
rlm@34 65 "Create a SceneProcessor object which wraps a vision processing
rlm@113 66 continuation function. The continuation is a function that takes
rlm@113 67 [#^Renderer r #^FrameBuffer fb #^ByteBuffer b #^BufferedImage bi],
rlm@306 68 each of which has already been appropriately sized."
rlm@23 69 [continuation]
rlm@23 70 (let [byte-buffer (atom nil)
rlm@113 71 renderer (atom nil)
rlm@113 72 image (atom nil)]
rlm@23 73 (proxy [SceneProcessor] []
rlm@23 74 (initialize
rlm@23 75 [renderManager viewPort]
rlm@23 76 (let [cam (.getCamera viewPort)
rlm@23 77 width (.getWidth cam)
rlm@23 78 height (.getHeight cam)]
rlm@23 79 (reset! renderer (.getRenderer renderManager))
rlm@23 80 (reset! byte-buffer
rlm@23 81 (BufferUtils/createByteBuffer
rlm@113 82 (* width height 4)))
rlm@113 83 (reset! image (BufferedImage.
rlm@113 84 width height
rlm@113 85 BufferedImage/TYPE_4BYTE_ABGR))))
rlm@23 86 (isInitialized [] (not (nil? @byte-buffer)))
rlm@23 87 (reshape [_ _ _])
rlm@23 88 (preFrame [_])
rlm@23 89 (postQueue [_])
rlm@23 90 (postFrame
rlm@23 91 [#^FrameBuffer fb]
rlm@23 92 (.clear @byte-buffer)
rlm@113 93 (continuation @renderer fb @byte-buffer @image))
rlm@23 94 (cleanup []))))
rlm@213 95 #+end_src
rlm@213 96
rlm@273 97 The continuation function given to =vision-pipeline= above will be
rlm@213 98 given a =Renderer= and three containers for image data. The
rlm@218 99 =FrameBuffer= references the GPU image data, but the pixel data can
rlm@218 100 not be used directly on the CPU. The =ByteBuffer= and =BufferedImage=
rlm@219 101 are initially "empty" but are sized to hold the data in the
rlm@306 102 =FrameBuffer=. I call transferring the GPU image data to the CPU
rlm@213 103 structures "mixing" the image data. I have provided three functions to
rlm@213 104 do this mixing.
rlm@213 105
rlm@213 106 #+name: pipeline-2
rlm@213 107 #+begin_src clojure
rlm@113 108 (defn frameBuffer->byteBuffer!
rlm@113 109 "Transfer the data in the graphics card (Renderer, FrameBuffer) to
rlm@113 110 the CPU (ByteBuffer)."
rlm@113 111 [#^Renderer r #^FrameBuffer fb #^ByteBuffer bb]
rlm@113 112 (.readFrameBuffer r fb bb) bb)
rlm@113 113
rlm@113 114 (defn byteBuffer->bufferedImage!
rlm@113 115 "Convert the C-style BGRA image data in the ByteBuffer bb to the AWT
rlm@113 116 style ABGR image data and place it in BufferedImage bi."
rlm@113 117 [#^ByteBuffer bb #^BufferedImage bi]
rlm@113 118 (Screenshots/convertScreenShot bb bi) bi)
rlm@113 119
rlm@113 120 (defn BufferedImage!
rlm@113 121 "Continuation which will grab the buffered image from the materials
rlm@113 122 provided by (vision-pipeline)."
rlm@113 123 [#^Renderer r #^FrameBuffer fb #^ByteBuffer bb #^BufferedImage bi]
rlm@113 124 (byteBuffer->bufferedImage!
rlm@113 125 (frameBuffer->byteBuffer! r fb bb) bi))
rlm@213 126 #+end_src
rlm@112 127
rlm@213 128 Note that it is possible to write vision processing algorithms
rlm@213 129 entirely in terms of =BufferedImage= inputs. Just compose that
rlm@273 130 =BufferedImage= algorithm with =BufferedImage!=. However, a vision
rlm@213 131 processing algorithm that is entirely hosted on the GPU does not have
rlm@306 132 to pay for this convenience.
rlm@213 133
ocsenave@265 134 * Optical sensor arrays are described with images and referenced with metadata
rlm@214 135 The vision pipeline described above handles the flow of rendered
rlm@214 136 images. Now, we need simulated eyes to serve as the source of these
rlm@214 137 images.
rlm@214 138
rlm@214 139 An eye is described in blender in the same way as a joint. They are
rlm@214 140 zero dimensional empty objects with no geometry whose local coordinate
rlm@214 141 system determines the orientation of the resulting eye. All eyes are
rlm@306 142 children of a parent node named "eyes" just as all joints have a
rlm@214 143 parent named "joints". An eye binds to the nearest physical object
rlm@273 144 with =bind-sense=.
rlm@214 145
rlm@214 146 #+name: add-eye
rlm@214 147 #+begin_src clojure
rlm@215 148 (in-ns 'cortex.vision)
rlm@215 149
rlm@214 150 (defn add-eye!
rlm@214 151 "Create a Camera centered on the current position of 'eye which
rlm@338 152 follows the closest physical node in 'creature. The camera will
rlm@338 153 point in the X direction and use the Z vector as up as determined
rlm@338 154 by the rotation of these vectors in blender coordinate space. Use
rlm@338 155 XZY rotation for the node in blender."
rlm@214 156 [#^Node creature #^Spatial eye]
rlm@214 157 (let [target (closest-node creature eye)
rlm@338 158 [cam-width cam-height]
rlm@338 159 ;;[640 480] ;; graphics card on laptop doesn't support
rlm@338 160 ;; arbitray dimensions.
rlm@338 161 (eye-dimensions eye)
rlm@215 162 cam (Camera. cam-width cam-height)
rlm@215 163 rot (.getWorldRotation eye)]
rlm@214 164 (.setLocation cam (.getWorldTranslation eye))
rlm@218 165 (.lookAtDirection
rlm@338 166 cam ; this part is not a mistake and
rlm@338 167 (.mult rot Vector3f/UNIT_X) ; is consistent with using Z in
rlm@338 168 (.mult rot Vector3f/UNIT_Y)) ; blender as the UP vector.
rlm@214 169 (.setFrustumPerspective
rlm@338 170 cam (float 45)
rlm@338 171 (float (/ (.getWidth cam) (.getHeight cam)))
rlm@338 172 (float 1)
rlm@338 173 (float 1000))
rlm@215 174 (bind-sense target cam) cam))
rlm@214 175 #+end_src
rlm@214 176
rlm@338 177 #+results: add-eye
rlm@338 178 : #'cortex.vision/add-eye!
rlm@338 179
rlm@214 180 Here, the camera is created based on metadata on the eye-node and
rlm@273 181 attached to the nearest physical object with =bind-sense=
rlm@214 182 ** The Retina
rlm@214 183
rlm@214 184 An eye is a surface (the retina) which contains many discrete sensors
rlm@218 185 to detect light. These sensors have can have different light-sensing
rlm@214 186 properties. In humans, each discrete sensor is sensitive to red,
rlm@214 187 blue, green, or gray. These different types of sensors can have
rlm@214 188 different spatial distributions along the retina. In humans, there is
rlm@214 189 a fovea in the center of the retina which has a very high density of
rlm@214 190 color sensors, and a blind spot which has no sensors at all. Sensor
rlm@219 191 density decreases in proportion to distance from the fovea.
rlm@214 192
rlm@214 193 I want to be able to model any retinal configuration, so my eye-nodes
rlm@214 194 in blender contain metadata pointing to images that describe the
rlm@306 195 precise position of the individual sensors using white pixels. The
rlm@306 196 meta-data also describes the precise sensitivity to light that the
rlm@214 197 sensors described in the image have. An eye can contain any number of
rlm@214 198 these images. For example, the metadata for an eye might look like
rlm@214 199 this:
rlm@214 200
rlm@214 201 #+begin_src clojure
rlm@214 202 {0xFF0000 "Models/test-creature/retina-small.png"}
rlm@214 203 #+end_src
rlm@214 204
rlm@214 205 #+caption: The retinal profile image "Models/test-creature/retina-small.png". White pixels are photo-sensitive elements. The distribution of white pixels is denser in the middle and falls off at the edges and is inspired by the human retina.
rlm@214 206 [[../assets/Models/test-creature/retina-small.png]]
rlm@214 207
rlm@214 208 Together, the number 0xFF0000 and the image image above describe the
rlm@214 209 placement of red-sensitive sensory elements.
rlm@214 210
rlm@214 211 Meta-data to very crudely approximate a human eye might be something
rlm@214 212 like this:
rlm@214 213
rlm@214 214 #+begin_src clojure
rlm@214 215 (let [retinal-profile "Models/test-creature/retina-small.png"]
rlm@214 216 {0xFF0000 retinal-profile
rlm@214 217 0x00FF00 retinal-profile
rlm@214 218 0x0000FF retinal-profile
rlm@214 219 0xFFFFFF retinal-profile})
rlm@214 220 #+end_src
rlm@214 221
rlm@214 222 The numbers that serve as keys in the map determine a sensor's
rlm@214 223 relative sensitivity to the channels red, green, and blue. These
rlm@218 224 sensitivity values are packed into an integer in the order =|_|R|G|B|=
rlm@218 225 in 8-bit fields. The RGB values of a pixel in the image are added
rlm@306 226 together with these sensitivities as linear weights. Therefore,
rlm@214 227 0xFF0000 means sensitive to red only while 0xFFFFFF means sensitive to
rlm@214 228 all colors equally (gray).
rlm@214 229
rlm@306 230 For convenience I've defined a few symbols for the more common
rlm@214 231 sensitivity values.
rlm@214 232
rlm@214 233 #+name: sensitivity
rlm@214 234 #+begin_src clojure
rlm@317 235 (def sensitivity-presets
rlm@317 236 "Retinal sensitivity presets for sensors that extract one channel
rlm@317 237 (:red :blue :green) or average all channels (:all)"
rlm@214 238 {:all 0xFFFFFF
rlm@214 239 :red 0xFF0000
rlm@214 240 :blue 0x0000FF
rlm@317 241 :green 0x00FF00})
rlm@214 242 #+end_src
rlm@214 243
rlm@214 244 ** Metadata Processing
rlm@214 245
rlm@273 246 =retina-sensor-profile= extracts a map from the eye-node in the same
rlm@273 247 format as the example maps above. =eye-dimensions= finds the
rlm@219 248 dimensions of the smallest image required to contain all the retinal
rlm@214 249 sensor maps.
rlm@214 250
rlm@216 251 #+name: retina
rlm@214 252 #+begin_src clojure
rlm@214 253 (defn retina-sensor-profile
rlm@214 254 "Return a map of pixel sensitivity numbers to BufferedImages
rlm@214 255 describing the distribution of light-sensitive components of this
rlm@214 256 eye. :red, :green, :blue, :gray are already defined as extracting
rlm@214 257 the red, green, blue, and average components respectively."
rlm@214 258 [#^Spatial eye]
rlm@214 259 (if-let [eye-map (meta-data eye "eye")]
rlm@214 260 (map-vals
rlm@214 261 load-image
rlm@214 262 (eval (read-string eye-map)))))
rlm@214 263
rlm@218 264 (defn eye-dimensions
rlm@218 265 "Returns [width, height] determined by the metadata of the eye."
rlm@214 266 [#^Spatial eye]
rlm@214 267 (let [dimensions
rlm@214 268 (map #(vector (.getWidth %) (.getHeight %))
rlm@214 269 (vals (retina-sensor-profile eye)))]
rlm@214 270 [(apply max (map first dimensions))
rlm@214 271 (apply max (map second dimensions))]))
rlm@214 272 #+end_src
rlm@214 273
ocsenave@265 274 * Importing and parsing descriptions of eyes.
rlm@214 275 First off, get the children of the "eyes" empty node to find all the
rlm@214 276 eyes the creature has.
rlm@216 277 #+name: eye-node
rlm@214 278 #+begin_src clojure
rlm@317 279 (def
rlm@317 280 ^{:doc "Return the children of the creature's \"eyes\" node."
rlm@317 281 :arglists '([creature])}
rlm@214 282 eyes
rlm@317 283 (sense-nodes "eyes"))
rlm@214 284 #+end_src
rlm@214 285
rlm@273 286 Then, add the camera created by =add-eye!= to the simulation by
rlm@215 287 creating a new viewport.
rlm@214 288
rlm@216 289 #+name: add-camera
rlm@213 290 #+begin_src clojure
rlm@338 291 (in-ns 'cortex.vision)
rlm@169 292 (defn add-camera!
rlm@169 293 "Add a camera to the world, calling continuation on every frame
rlm@34 294 produced."
rlm@167 295 [#^Application world camera continuation]
rlm@23 296 (let [width (.getWidth camera)
rlm@23 297 height (.getHeight camera)
rlm@23 298 render-manager (.getRenderManager world)
rlm@23 299 viewport (.createMainView render-manager "eye-view" camera)]
rlm@23 300 (doto viewport
rlm@23 301 (.setClearFlags true true true)
rlm@112 302 (.setBackgroundColor ColorRGBA/Black)
rlm@113 303 (.addProcessor (vision-pipeline continuation))
rlm@23 304 (.attachScene (.getRootNode world)))))
rlm@215 305 #+end_src
rlm@151 306
rlm@338 307 #+results: add-camera
rlm@338 308 : #'cortex.vision/add-camera!
rlm@338 309
rlm@151 310
rlm@218 311 The eye's continuation function should register the viewport with the
rlm@218 312 simulation the first time it is called, use the CPU to extract the
rlm@215 313 appropriate pixels from the rendered image and weight them by each
rlm@218 314 sensor's sensitivity. I have the option to do this processing in
rlm@218 315 native code for a slight gain in speed. I could also do it in the GPU
rlm@273 316 for a massive gain in speed. =vision-kernel= generates a list of
rlm@218 317 such continuation functions, one for each channel of the eye.
rlm@151 318
rlm@216 319 #+name: kernel
rlm@215 320 #+begin_src clojure
rlm@215 321 (in-ns 'cortex.vision)
rlm@151 322
rlm@215 323 (defrecord attached-viewport [vision-fn viewport-fn]
rlm@215 324 clojure.lang.IFn
rlm@215 325 (invoke [this world] (vision-fn world))
rlm@215 326 (applyTo [this args] (apply vision-fn args)))
rlm@151 327
rlm@216 328 (defn pixel-sense [sensitivity pixel]
rlm@216 329 (let [s-r (bit-shift-right (bit-and 0xFF0000 sensitivity) 16)
rlm@216 330 s-g (bit-shift-right (bit-and 0x00FF00 sensitivity) 8)
rlm@216 331 s-b (bit-and 0x0000FF sensitivity)
rlm@216 332
rlm@216 333 p-r (bit-shift-right (bit-and 0xFF0000 pixel) 16)
rlm@216 334 p-g (bit-shift-right (bit-and 0x00FF00 pixel) 8)
rlm@216 335 p-b (bit-and 0x0000FF pixel)
rlm@216 336
rlm@216 337 total-sensitivity (* 255 (+ s-r s-g s-b))]
rlm@216 338 (float (/ (+ (* s-r p-r)
rlm@216 339 (* s-g p-g)
rlm@216 340 (* s-b p-b))
rlm@216 341 total-sensitivity))))
rlm@216 342
rlm@215 343 (defn vision-kernel
rlm@171 344 "Returns a list of functions, each of which will return a color
rlm@171 345 channel's worth of visual information when called inside a running
rlm@171 346 simulation."
rlm@151 347 [#^Node creature #^Spatial eye & {skip :skip :or {skip 0}}]
rlm@169 348 (let [retinal-map (retina-sensor-profile eye)
rlm@169 349 camera (add-eye! creature eye)
rlm@151 350 vision-image
rlm@151 351 (atom
rlm@151 352 (BufferedImage. (.getWidth camera)
rlm@151 353 (.getHeight camera)
rlm@170 354 BufferedImage/TYPE_BYTE_BINARY))
rlm@170 355 register-eye!
rlm@170 356 (runonce
rlm@170 357 (fn [world]
rlm@170 358 (add-camera!
rlm@170 359 world camera
rlm@170 360 (let [counter (atom 0)]
rlm@170 361 (fn [r fb bb bi]
rlm@170 362 (if (zero? (rem (swap! counter inc) (inc skip)))
rlm@170 363 (reset! vision-image
rlm@170 364 (BufferedImage! r fb bb bi))))))))]
rlm@151 365 (vec
rlm@151 366 (map
rlm@151 367 (fn [[key image]]
rlm@151 368 (let [whites (white-coordinates image)
rlm@151 369 topology (vec (collapse whites))
rlm@216 370 sensitivity (sensitivity-presets key key)]
rlm@215 371 (attached-viewport.
rlm@215 372 (fn [world]
rlm@215 373 (register-eye! world)
rlm@215 374 (vector
rlm@215 375 topology
rlm@215 376 (vec
rlm@215 377 (for [[x y] whites]
rlm@216 378 (pixel-sense
rlm@216 379 sensitivity
rlm@216 380 (.getRGB @vision-image x y))))))
rlm@215 381 register-eye!)))
rlm@215 382 retinal-map))))
rlm@151 383
rlm@215 384 (defn gen-fix-display
rlm@215 385 "Create a function to call to restore a simulation's display when it
rlm@215 386 is disrupted by a Viewport."
rlm@215 387 []
rlm@215 388 (runonce
rlm@215 389 (fn [world]
rlm@215 390 (add-camera! world (.getCamera world) no-op))))
rlm@215 391 #+end_src
rlm@170 392
rlm@273 393 Note that since each of the functions generated by =vision-kernel=
rlm@273 394 shares the same =register-eye!= function, the eye will be registered
rlm@215 395 only once the first time any of the functions from the list returned
rlm@273 396 by =vision-kernel= is called. Each of the functions returned by
rlm@273 397 =vision-kernel= also allows access to the =Viewport= through which
rlm@306 398 it receives images.
rlm@215 399
rlm@306 400 The in-game display can be disrupted by all the ViewPorts that the
rlm@306 401 functions generated by =vision-kernel= add. This doesn't affect the
rlm@215 402 simulation or the simulated senses, but can be annoying.
rlm@273 403 =gen-fix-display= restores the in-simulation display.
rlm@215 404
ocsenave@265 405 ** The =vision!= function creates sensory probes.
rlm@215 406
rlm@218 407 All the hard work has been done; all that remains is to apply
rlm@273 408 =vision-kernel= to each eye in the creature and gather the results
rlm@215 409 into one list of functions.
rlm@215 410
rlm@216 411 #+name: main
rlm@215 412 #+begin_src clojure
rlm@170 413 (defn vision!
rlm@348 414 "Returns a list of functions, each of which returns visual sensory
rlm@348 415 data when called inside a running simulation."
rlm@151 416 [#^Node creature & {skip :skip :or {skip 0}}]
rlm@151 417 (reduce
rlm@170 418 concat
rlm@167 419 (for [eye (eyes creature)]
rlm@215 420 (vision-kernel creature eye))))
rlm@215 421 #+end_src
rlm@151 422
ocsenave@265 423 ** Displaying visual data for debugging.
ocsenave@265 424 # Visualization of Vision. Maybe less alliteration would be better.
rlm@215 425 It's vital to have a visual representation for each sense. Here I use
rlm@273 426 =view-sense= to construct a function that will create a display for
rlm@215 427 visual data.
rlm@215 428
rlm@216 429 #+name: display
rlm@215 430 #+begin_src clojure
rlm@216 431 (in-ns 'cortex.vision)
rlm@216 432
rlm@189 433 (defn view-vision
rlm@189 434 "Creates a function which accepts a list of visual sensor-data and
rlm@189 435 displays each element of the list to the screen."
rlm@189 436 []
rlm@188 437 (view-sense
rlm@188 438 (fn
rlm@188 439 [[coords sensor-data]]
rlm@188 440 (let [image (points->image coords)]
rlm@188 441 (dorun
rlm@188 442 (for [i (range (count coords))]
rlm@188 443 (.setRGB image ((coords i) 0) ((coords i) 1)
rlm@216 444 (gray (int (* 255 (sensor-data i)))))))
rlm@189 445 image))))
rlm@34 446 #+end_src
rlm@23 447
ocsenave@264 448 * Demonstrations
ocsenave@264 449 ** Demonstrating the vision pipeline.
rlm@23 450
rlm@215 451 This is a basic test for the vision system. It only tests the
ocsenave@264 452 vision-pipeline and does not deal with loading eyes from a blender
rlm@215 453 file. The code creates two videos of the same rotating cube from
rlm@215 454 different angles.
rlm@23 455
rlm@215 456 #+name: test-1
rlm@23 457 #+begin_src clojure
rlm@215 458 (in-ns 'cortex.test.vision)
rlm@23 459
rlm@219 460 (defn test-pipeline
rlm@69 461 "Testing vision:
rlm@69 462 Tests the vision system by creating two views of the same rotating
rlm@69 463 object from different angles and displaying both of those views in
rlm@69 464 JFrames.
rlm@69 465
rlm@69 466 You should see a rotating cube, and two windows,
rlm@69 467 each displaying a different view of the cube."
rlm@283 468 ([] (test-pipeline false))
rlm@283 469 ([record?]
rlm@283 470 (let [candy
rlm@283 471 (box 1 1 1 :physical? false :color ColorRGBA/Blue)]
rlm@283 472 (world
rlm@283 473 (doto (Node.)
rlm@283 474 (.attachChild candy))
rlm@283 475 {}
rlm@283 476 (fn [world]
rlm@283 477 (let [cam (.clone (.getCamera world))
rlm@283 478 width (.getWidth cam)
rlm@283 479 height (.getHeight cam)]
rlm@283 480 (add-camera! world cam
rlm@283 481 (comp
rlm@283 482 (view-image
rlm@283 483 (if record?
rlm@283 484 (File. "/home/r/proj/cortex/render/vision/1")))
rlm@283 485 BufferedImage!))
rlm@283 486 (add-camera! world
rlm@283 487 (doto (.clone cam)
rlm@283 488 (.setLocation (Vector3f. -10 0 0))
rlm@283 489 (.lookAt Vector3f/ZERO Vector3f/UNIT_Y))
rlm@283 490 (comp
rlm@283 491 (view-image
rlm@283 492 (if record?
rlm@283 493 (File. "/home/r/proj/cortex/render/vision/2")))
rlm@283 494 BufferedImage!))
rlm@341 495 (let [timer (IsoTimer. 60)]
rlm@340 496 (.setTimer world timer)
rlm@340 497 (display-dilated-time world timer))
rlm@283 498 ;; This is here to restore the main view
rlm@340 499 ;; after the other views have completed processing
rlm@283 500 (add-camera! world (.getCamera world) no-op)))
rlm@283 501 (fn [world tpf]
rlm@283 502 (.rotate candy (* tpf 0.2) 0 0))))))
rlm@23 503 #+end_src
rlm@23 504
rlm@340 505 #+results: test-1
rlm@340 506 : #'cortex.test.vision/test-pipeline
rlm@340 507
rlm@215 508 #+begin_html
rlm@215 509 <div class="figure">
rlm@215 510 <video controls="controls" width="755">
rlm@215 511 <source src="../video/spinning-cube.ogg" type="video/ogg"
rlm@215 512 preload="none" poster="../images/aurellem-1280x480.png" />
rlm@215 513 </video>
rlm@309 514 <br> <a href="http://youtu.be/r5Bn2aG7MO0"> YouTube </a>
rlm@215 515 <p>A rotating cube viewed from two different perspectives.</p>
rlm@215 516 </div>
rlm@215 517 #+end_html
rlm@215 518
rlm@215 519 Creating multiple eyes like this can be used for stereoscopic vision
rlm@215 520 simulation in a single creature or for simulating multiple creatures,
rlm@215 521 each with their own sense of vision.
ocsenave@264 522 ** Demonstrating eye import and parsing.
rlm@215 523
rlm@218 524 To the worm from the last post, I add a new node that describes its
rlm@215 525 eyes.
rlm@215 526
rlm@215 527 #+attr_html: width=755
rlm@215 528 #+caption: The worm with newly added empty nodes describing a single eye.
rlm@215 529 [[../images/worm-with-eye.png]]
rlm@215 530
rlm@215 531 The node highlighted in yellow is the root level "eyes" node. It has
rlm@218 532 a single child, highlighted in orange, which describes a single
rlm@218 533 eye. This is the "eye" node. It is placed so that the worm will have
rlm@218 534 an eye located in the center of the flat portion of its lower
rlm@218 535 hemispherical section.
rlm@218 536
rlm@218 537 The two nodes which are not highlighted describe the single joint of
rlm@218 538 the worm.
rlm@215 539
rlm@215 540 The metadata of the eye-node is:
rlm@215 541
rlm@215 542 #+begin_src clojure :results verbatim :exports both
rlm@215 543 (cortex.sense/meta-data
rlm@218 544 (.getChild (.getChild (cortex.test.body/worm) "eyes") "eye") "eye")
rlm@215 545 #+end_src
rlm@215 546
rlm@215 547 #+results:
rlm@215 548 : "(let [retina \"Models/test-creature/retina-small.png\"]
rlm@215 549 : {:all retina :red retina :green retina :blue retina})"
rlm@215 550
rlm@215 551 This is the approximation to the human eye described earlier.
rlm@215 552
rlm@216 553 #+name: test-2
rlm@215 554 #+begin_src clojure
rlm@215 555 (in-ns 'cortex.test.vision)
rlm@215 556
rlm@216 557 (defn change-color [obj color]
rlm@321 558 ;;(println-repl obj)
rlm@216 559 (if obj
rlm@216 560 (.setColor (.getMaterial obj) "Color" color)))
rlm@216 561
rlm@216 562 (defn colored-cannon-ball [color]
rlm@216 563 (comp #(change-color % color)
rlm@216 564 (fire-cannon-ball)))
rlm@215 565
rlm@338 566 (defn gen-worm
rlm@338 567 "create a creature acceptable for testing as a replacement for the
rlm@338 568 worm."
rlm@338 569 []
rlm@338 570 (nodify
rlm@338 571 "worm"
rlm@338 572 [(nodify
rlm@338 573 "eyes"
rlm@338 574 [(doto
rlm@338 575 (Node. "eye1")
rlm@338 576 (.setLocalTranslation (Vector3f. 0 -1.1 0))
rlm@338 577 (.setUserData
rlm@338 578
rlm@338 579 "eye"
rlm@338 580 "(let [retina
rlm@338 581 \"Models/test-creature/retina-small.png\"]
rlm@338 582 {:all retina :red retina
rlm@338 583 :green retina :blue retina})"))])
rlm@338 584 (box
rlm@338 585 0.2 0.2 0.2
rlm@338 586 :name "worm-segment"
rlm@338 587 :position (Vector3f. 0 0 0)
rlm@338 588 :color ColorRGBA/Orange)]))
rlm@338 589
rlm@338 590
rlm@338 591
rlm@283 592 (defn test-worm-vision
rlm@321 593 "Testing vision:
rlm@321 594 You should see the worm suspended in mid-air, looking down at a
rlm@321 595 table. There are four small displays, one each for red, green blue,
rlm@321 596 and gray channels. You can fire balls of various colors, and the
rlm@321 597 four channels should react accordingly.
rlm@321 598
rlm@321 599 Keys:
rlm@321 600 r : fire red-ball
rlm@321 601 b : fire blue-ball
rlm@321 602 g : fire green-ball
rlm@321 603 <space> : fire white ball"
rlm@338 604
rlm@283 605 ([] (test-worm-vision false))
rlm@283 606 ([record?]
rlm@283 607 (let [the-worm (doto (worm)(body!))
rlm@340 608 vision (vision! the-worm)
rlm@340 609 vision-display (view-vision)
rlm@340 610 fix-display (gen-fix-display)
rlm@283 611 me (sphere 0.5 :color ColorRGBA/Blue :physical? false)
rlm@283 612 x-axis
rlm@283 613 (box 1 0.01 0.01 :physical? false :color ColorRGBA/Red
rlm@283 614 :position (Vector3f. 0 -5 0))
rlm@283 615 y-axis
rlm@283 616 (box 0.01 1 0.01 :physical? false :color ColorRGBA/Green
rlm@283 617 :position (Vector3f. 0 -5 0))
rlm@283 618 z-axis
rlm@283 619 (box 0.01 0.01 1 :physical? false :color ColorRGBA/Blue
rlm@283 620 :position (Vector3f. 0 -5 0))
rlm@340 621
rlm@338 622 ]
rlm@215 623
rlm@335 624 (world
rlm@335 625 (nodify [(floor) the-worm x-axis y-axis z-axis me])
rlm@340 626 (merge standard-debug-controls
rlm@340 627 {"key-r" (colored-cannon-ball ColorRGBA/Red)
rlm@340 628 "key-b" (colored-cannon-ball ColorRGBA/Blue)
rlm@340 629 "key-g" (colored-cannon-ball ColorRGBA/Green)})
rlm@338 630
rlm@335 631 (fn [world]
rlm@340 632 (light-up-everything world)
rlm@340 633 (speed-up world)
rlm@341 634 (let [timer (IsoTimer. 60)]
rlm@340 635 (.setTimer world timer)
rlm@340 636 (display-dilated-time world timer))
rlm@340 637 ;; add a view from the worm's perspective
rlm@340 638 (if record?
rlm@340 639 (Capture/captureVideo
rlm@340 640 world
rlm@340 641 (File.
rlm@340 642 "/home/r/proj/cortex/render/worm-vision/main-view")))
rlm@340 643
rlm@340 644 (add-camera!
rlm@340 645 world
rlm@340 646 (add-eye! the-worm (first (eyes the-worm)))
rlm@340 647 (comp
rlm@340 648 (view-image
rlm@340 649 (if record?
rlm@340 650 (File.
rlm@340 651 "/home/r/proj/cortex/render/worm-vision/worm-view")))
rlm@340 652 BufferedImage!))
rlm@340 653
rlm@340 654 (set-gravity world Vector3f/ZERO)
rlm@340 655 (add-camera! world (.getCamera world) no-op))
rlm@340 656
rlm@340 657 (fn [world _]
rlm@340 658 (.setLocalTranslation me (.getLocation (.getCamera world)))
rlm@340 659 (vision-display
rlm@340 660 (map #(% world) vision)
rlm@338 661 (if record?
rlm@340 662 (File. "/home/r/proj/cortex/render/worm-vision")))
rlm@340 663 (fix-display world)
rlm@335 664 )))))
rlm@215 665 #+end_src
rlm@215 666
rlm@335 667 #+RESULTS: test-2
rlm@337 668 : #'cortex.test.vision/test-worm-vision
rlm@335 669
rlm@335 670
rlm@218 671 The world consists of the worm and a flat gray floor. I can shoot red,
rlm@218 672 green, blue and white cannonballs at the worm. The worm is initially
rlm@218 673 looking down at the floor, and there is no gravity. My perspective
rlm@218 674 (the Main View), the worm's perspective (Worm View) and the 4 sensor
rlm@218 675 channels that comprise the worm's eye are all saved frame-by-frame to
rlm@218 676 disk.
rlm@218 677
rlm@218 678 * Demonstration of Vision
rlm@218 679 #+begin_html
rlm@218 680 <div class="figure">
rlm@218 681 <video controls="controls" width="755">
rlm@218 682 <source src="../video/worm-vision.ogg" type="video/ogg"
rlm@218 683 preload="none" poster="../images/aurellem-1280x480.png" />
rlm@218 684 </video>
rlm@309 685 <br> <a href="http://youtu.be/J3H3iB_2NPQ"> YouTube </a>
rlm@218 686 <p>Simulated Vision in a Virtual Environment</p>
rlm@218 687 </div>
rlm@218 688 #+end_html
rlm@218 689
rlm@218 690 ** Generate the Worm Video from Frames
rlm@216 691 #+name: magick2
rlm@216 692 #+begin_src clojure
rlm@216 693 (ns cortex.video.magick2
rlm@216 694 (:import java.io.File)
rlm@316 695 (:use clojure.java.shell))
rlm@216 696
rlm@216 697 (defn images [path]
rlm@216 698 (sort (rest (file-seq (File. path)))))
rlm@216 699
rlm@216 700 (def base "/home/r/proj/cortex/render/worm-vision/")
rlm@216 701
rlm@216 702 (defn pics [file]
rlm@216 703 (images (str base file)))
rlm@216 704
rlm@216 705 (defn combine-images []
rlm@216 706 (let [main-view (pics "main-view")
rlm@216 707 worm-view (pics "worm-view")
rlm@216 708 blue (pics "0")
rlm@216 709 green (pics "1")
rlm@216 710 red (pics "2")
rlm@216 711 gray (pics "3")
rlm@216 712 blender (let [b-pics (pics "blender")]
rlm@216 713 (concat b-pics (repeat 9001 (last b-pics))))
rlm@216 714 background (repeat 9001 (File. (str base "background.png")))
rlm@216 715 targets (map
rlm@216 716 #(File. (str base "out/" (format "%07d.png" %)))
rlm@216 717 (range 0 (count main-view)))]
rlm@216 718 (dorun
rlm@216 719 (pmap
rlm@216 720 (comp
rlm@216 721 (fn [[background main-view worm-view red green blue gray blender target]]
rlm@216 722 (println target)
rlm@216 723 (sh "convert"
rlm@216 724 background
rlm@216 725 main-view "-geometry" "+18+17" "-composite"
rlm@216 726 worm-view "-geometry" "+677+17" "-composite"
rlm@216 727 green "-geometry" "+685+430" "-composite"
rlm@216 728 red "-geometry" "+788+430" "-composite"
rlm@216 729 blue "-geometry" "+894+430" "-composite"
rlm@216 730 gray "-geometry" "+1000+430" "-composite"
rlm@216 731 blender "-geometry" "+0+0" "-composite"
rlm@216 732 target))
rlm@216 733 (fn [& args] (map #(.getCanonicalPath %) args)))
rlm@216 734 background main-view worm-view red green blue gray blender targets))))
rlm@216 735 #+end_src
rlm@216 736
rlm@216 737 #+begin_src sh :results silent
rlm@216 738 cd /home/r/proj/cortex/render/worm-vision
rlm@216 739 ffmpeg -r 25 -b 9001k -i out/%07d.png -vcodec libtheora worm-vision.ogg
rlm@216 740 #+end_src
rlm@236 741
ocsenave@265 742 * Onward!
ocsenave@265 743 - As a neat bonus, this idea behind simulated vision also enables one
ocsenave@265 744 to [[../../cortex/html/capture-video.html][capture live video feeds from jMonkeyEngine]].
ocsenave@265 745 - Now that we have vision, it's time to tackle [[./hearing.org][hearing]].
ocsenave@265 746 #+appendix
ocsenave@265 747
rlm@215 748 * Headers
rlm@215 749
rlm@213 750 #+name: vision-header
rlm@213 751 #+begin_src clojure
rlm@213 752 (ns cortex.vision
rlm@213 753 "Simulate the sense of vision in jMonkeyEngine3. Enables multiple
rlm@213 754 eyes from different positions to observe the same world, and pass
rlm@306 755 the observed data to any arbitrary function. Automatically reads
rlm@216 756 eye-nodes from specially prepared blender files and instantiates
rlm@213 757 them in the world as actual eyes."
rlm@213 758 {:author "Robert McIntyre"}
rlm@213 759 (:use (cortex world sense util))
rlm@213 760 (:import com.jme3.post.SceneProcessor)
rlm@237 761 (:import (com.jme3.util BufferUtils Screenshots))
rlm@213 762 (:import java.nio.ByteBuffer)
rlm@213 763 (:import java.awt.image.BufferedImage)
rlm@213 764 (:import (com.jme3.renderer ViewPort Camera))
rlm@216 765 (:import (com.jme3.math ColorRGBA Vector3f Matrix3f))
rlm@213 766 (:import com.jme3.renderer.Renderer)
rlm@213 767 (:import com.jme3.app.Application)
rlm@213 768 (:import com.jme3.texture.FrameBuffer)
rlm@213 769 (:import (com.jme3.scene Node Spatial)))
rlm@213 770 #+end_src
rlm@112 771
rlm@215 772 #+name: test-header
rlm@215 773 #+begin_src clojure
rlm@215 774 (ns cortex.test.vision
rlm@215 775 (:use (cortex world sense util body vision))
rlm@215 776 (:use cortex.test.body)
rlm@215 777 (:import java.awt.image.BufferedImage)
rlm@215 778 (:import javax.swing.JPanel)
rlm@215 779 (:import javax.swing.SwingUtilities)
rlm@215 780 (:import java.awt.Dimension)
rlm@215 781 (:import javax.swing.JFrame)
rlm@215 782 (:import com.jme3.math.ColorRGBA)
rlm@215 783 (:import com.jme3.scene.Node)
rlm@215 784 (:import com.jme3.math.Vector3f)
rlm@216 785 (:import java.io.File)
rlm@341 786 (:import (com.aurellem.capture Capture RatchetTimer IsoTimer)))
rlm@215 787 #+end_src
rlm@341 788
rlm@341 789 #+results: test-header
rlm@341 790 : com.aurellem.capture.IsoTimer
rlm@341 791
rlm@216 792 * Source Listing
rlm@216 793 - [[../src/cortex/vision.clj][cortex.vision]]
rlm@216 794 - [[../src/cortex/test/vision.clj][cortex.test.vision]]
rlm@216 795 - [[../src/cortex/video/magick2.clj][cortex.video.magick2]]
rlm@216 796 - [[../assets/Models/subtitles/worm-vision-subtitles.blend][worm-vision-subtitles.blend]]
rlm@216 797 #+html: <ul> <li> <a href="../org/sense.org">This org file</a> </li> </ul>
rlm@216 798 - [[http://hg.bortreb.com ][source-repository]]
rlm@216 799
rlm@35 800
rlm@273 801 * Next
rlm@273 802 I find some [[./hearing.org][ears]] for the creature while exploring the guts of
rlm@273 803 jMonkeyEngine's sound system.
rlm@24 804
rlm@212 805 * COMMENT Generate Source
rlm@34 806 #+begin_src clojure :tangle ../src/cortex/vision.clj
rlm@216 807 <<vision-header>>
rlm@216 808 <<pipeline-1>>
rlm@216 809 <<pipeline-2>>
rlm@216 810 <<retina>>
rlm@216 811 <<add-eye>>
rlm@216 812 <<sensitivity>>
rlm@216 813 <<eye-node>>
rlm@216 814 <<add-camera>>
rlm@216 815 <<kernel>>
rlm@216 816 <<main>>
rlm@216 817 <<display>>
rlm@24 818 #+end_src
rlm@24 819
rlm@68 820 #+begin_src clojure :tangle ../src/cortex/test/vision.clj
rlm@215 821 <<test-header>>
rlm@215 822 <<test-1>>
rlm@216 823 <<test-2>>
rlm@24 824 #+end_src
rlm@216 825
rlm@216 826 #+begin_src clojure :tangle ../src/cortex/video/magick2.clj
rlm@216 827 <<magick2>>
rlm@216 828 #+end_src