视觉认知的边界：《观察方式》解析计算机视觉目标感知

需积分: 7 26 浏览量更新于2024-07-18 收藏 3.04MB PDF 举报

《观察方式：视觉认知的范围和局限》(Ways of Seeing: The Scope and Limits of Visual Cognition)是由Pierre Jacob和Marc Jeannerod合著的一本书，它属于牛津认知科学系列的一部分，该系列由Martin Davies、James Higginbotham、Philip Johnson-Laird等人共同编辑。这本书主要探讨了视觉认知在计算机视觉领域的应用和理论基础，特别关注目标感知的方法，强调从视觉感知的角度深入理解这一复杂主题。书中涉及的核心概念包括"概念：认知科学的误区"（Concepts: Where Cognitive Science Went Wrong）系列中的几个专题，如Jerry A. Fodor关于上下文与内容的讨论、Robert C. Stalnaker的"心灵解读"（Mindreading）、Stephen Stich和Shaun Nichols的面部认知科学研究，以及Andy Young关于参照和意识的篇章。此外，书中的另一部分探讨了学习思考过程中的图像和语言，由John Campbell编写的"Seeing Reason: Image and Language in Learning to Think"提供了深入的见解。作者Pierre Jacob和Marc Jeannerod作为计算机视觉领域的专家，他们通过简洁明了的语言，引导读者理解视觉系统如何处理和解析我们日常生活中的视觉信息，同时也揭示了视觉认知的局限性，比如可能存在的偏见、误解或者信息处理的复杂性。他们可能还讨论了如何通过计算机技术来模拟或超越这些人类视觉机制，以及技术在提升人工智能视觉性能中的挑战。《观察方式》不仅是一本理论著作，也包含了实证研究和案例分析，让读者在了解视觉认知理论的同时，也能看到其在实际问题解决中的应用。这是一本对于视觉科学家、人工智能研究人员、心理学家以及对视觉感知感兴趣的所有专业人士来说不可多得的参考书籍。无论是在学术研究还是教育领域，都能找到有价值的信息和洞见。

that one is not able to conceptualize and the alleged distinctive link between the

fine-grained content of perceptual experience and the fine-tuning of bodily actions

directed towards objects. In agreement with other non-conceptualists, we both think

that visual perceptual experiences are more fine-grained and informationally richer than

thoughts and that a creature may enjoy visual experiences for which she has no corres-

ponding concepts. The dual model of human vision, however, is not consistent with the

alleged link between the non-conceptual content of visual experiences and the fine

control of action.

Since they insist that the conceptual content of thought and language fails to capture

the fine-grainedness of the non-conceptual content of visual experiences, it is natural for

non-conceptualist philosophers to assume that the non-conceptual content of visual

experiences must match the requirements of bodily actions. From a non-conceptualist

standpoint then, the natural temptation is to link the fine-grainedness of conscious visual

experiences to the fine-tuning of bodily movements. For example, consider Peacocke’s

notion of a ‘scenario content’ designed to capture part of the non-conceptual spatial

content of a visual percept. Peacocke (1992a: 93), who thinks of scenario content as ‘an

intermediary between perception and action’, argues that ‘in supplying a subject with

information about the location of things relative to bodily axes, perception supplies that

non-conceptual information in a form immediately usable if the subject wants to move

his body or some limb toward, from, or in some other spatial relation to what he per-

ceives’. Similarly, according to O’Shaughnessy (1992: 231–33), the task of conscious

visual experiences is to ‘assist’, or ‘cooperate’ with, one’s bodily movements: ‘sight

guides an instrumental act to its target’. The temptation is to assume that the richness

and fine-grainedness of sight or the non-conceptual content of visual experiences is at

the service of the fine attunement of bodily movements. A version of this temptation

has recently been expressed by Clark (2001) as what he calls ‘the assumption of

experience-based control’ (EBC), i.e. the view that ‘conscious visual experience

presents the world to the subject in a richly textured way; a way that presents fine detail

(detail that may [. . .] exceed our conceptual or propositional grasp) and that is, in virtue

of this richness, especially apt for, and typically utilized in, the control and guidance of

fine-tuned, real-world activity’.

What seems to lend prima facie support to some version of the thesis of ‘experience-

based control’ is the contrast between the computational requirements made upon the

human reasoning and memory systems about objects, and upon the system dedicated

to acting on objects, respectively. The job of the former is to encode more abstract

information about the enduring properties of objects in a format that is object-dependent

and viewer-independent (so as to ensure object-recognition from different subjective

viewpoints). The job of the latter is to keep track, and regularly update, the representa-

tions of the constantly changing features of visual objects, within an egocentric frame of

reference suitable for the guidance and monitoring of finely adjusted bodily movements.

We do agree that the fine-tuning of bodily movements involved in reaching and grasp-

ing objects does require the computation of visual information about objects different

from the visual information relevant for the purpose of categorizing objects, reasoning

and memorizing information about them. Indeed, one cannot visually reach and grasp

an object unless one’s visual system has computed the location of the object within an

egocentric frame of reference centered on the axis of one’s body. But we do not think

Introduction

Jacob-FM.qxd 8/23/03 4:55 PM Page xv

that the viewer-dependent perspective that is efficient for reaching and grasping objects

is an efficient system for memorizing information about an object’s properties so as to

maximize the likelihood of object-recognition over time.

In fact, the body of experimental research that supports the dual model of the human

visual processing of objects, shows that the human visuomotor transformation at work

in the guidance of visual actions directed towards objects is an automatic system that is

largely independent from the system that subserves the visual perceptual system. In the

everyday life of a normal human being, the visuomotor system and the visual

perceptual system collaborate and work in tandem. But brain lesions can disrupt this

cooperation. And so can carefully designed psychophysical experiments in normal

human subjects. Even normal daily life can provide numerous dissociations between

the fine-grainedness of conscious visual experience and the fine attunement of the auto-

matic control of bodily movements: the artist is immersed in the visual conscious experi-

ence of the shapes, colors and textures of the fruits she is copying. Her conscious focus

onto the visual experience of the fruits in no way interferes with her hand movement:

with the fingers of her right hand, she deftly grasps and swiftly lifts her brush onto

the canvas. Contrary to the assumption of experience-based control (EBC),

the non-conceptual content of visual experiences is not geared towards the guidance

and control of action. Rather, as Clark (2001) aptly puts it, it is geared towards the

‘selection’ of objects that can be either goals for visually guided actions or food for

thought. Nor does the alignment of the perceptual visual system with higher cognitive

functions (thought, reasoning, the belief-box and memory) force us to assume that the

content of visual perceptual experiences is itself conceptualized. The non-conceptual

content of visual experiences is poised for conceptualization. To say that it is poised

for conceptual use is not to say that it is conceptual content.

The structure of the book is as follows. Part I, entitled ‘The purposes of vision:

perceiving, thinking and acting’, is constituted by Chapter 1, in which we try to offer

a general representational framework for thinking about the puzzles of human vision.

There, we lay the groundwork for the rest of the book by sketching a teleosemantic

framework for thinking about the notion of mental representation, which, we think, is,

if not required by, at least consistent with, the practice of cognitive neuroscience. We

also examine the dynamical processes by which visual percepts give rise to more

detached thoughts and visuomotor representations feed motor intentions. We claim that

both visual percepts and visuomotor representations are mental representations with

non-conceptual content, but they differ significantly from one another. Visual illusions

provide an interesting example of the dissociation between the way things may visually

look and one’s considered judgment. The visual appearances are to some extent immune

to what one knows. Conversely, one’s considered judgment may diverge from the way

things look. Percepts provide no more than evidence for thoughts. Many recent experi-

ments have revealed a still further dissociation: one and the same visual stimulus can

give rise to a size-contrast illusory percept and to a different visuomotor representation

of a target of prehension, whose content differs in some important respects from the

content of the percept. In Chapter 4, we review these experiments. In Chapter 1, we also

discuss the arguments offered by philosophers on behalf of the distinction between the

conceptual content of thoughts and beliefs, and the non-conceptual content of perceptual

representations. As we emphasize, these arguments draw on the fine-grainedness and the

informational richness of the phenomenology of perceptual experiences. In a way, our

xvi

Introduction

Jacob-FM.qxd 8/23/03 4:55 PM Page xvi

book is an argument for the introduction of a new kind of non-conceptual content: the

content of visuomotor representations. The content of visuomotor representations is not

conceptual, but it is not perceptual either. As the book makes clear, the argument

for dividing non-conceptual content into perceptual and visuomotor content cannot, for

principled reasons, rely on the phenomenology of visual experience because, as we said

above, vision is broader than sight and, unlike visual perception, the visuomotor

processing of a target for action does not give rise to visual awareness of the world.

Part II, entitled ‘Empirical evidence for the duality of visual processing’, comprises

Chapters 2, 3 and 4. There, we review much experimental work ranging from

electrophysiological recordings of single cells in the brain of macaque monkeys and

behavioral experiments on animals to psychophysical experiments on normal human

subjects, through the neuropsychological study of human patients with a brain lesion in

their visual system. In Chapter 2, we contrast the responses of single neurons located,

respectively, in the anterior intraparietal area (AIP) and in the inferotemporal area (TE)

of the brain of macaque monkeys. The former respond to the geometrical properties of

objects relevant in the context of grasping tasks. The latter respond to more complex

visual properties of the same objects. In Chapter 3, we examine various degrees of

visual impairment caused by lesions in the ventral pathway and in the dorsal pathway

of the human visual system. In particular, we contrast visual-form agnosias and optic

ataxia. In Chapter 4, we review a rich body of psychophysical experiments in normal

human subjects that reveal many dissociations between perceptual and visuomotor pro-

cessing. In particular, we try to assess the complex experimental situation in the study

of some size-contrast visual illusions. Succinctly put, we argue that this rich and diverse

body of experimental work reveals that much visual processing in the human brain is

designed, not so much for the benefit of visual awareness and visual knowledge of the

world, as for the guidance of visual actions towards neighboring objects.

In Part III, entitled ‘Perceiving objects and grasping them’, constituted by Chapters 5

and 6, we try to provide a detailed analysis of the differences between visual percepts

and visuomotor representations. In Chapter 5, we examine the modularity of visual per-

cepts: visual percepts are so informationally encapsulated that some of the things that

can be thought cannot be perceived. We accept the distinction between non-epistemic

and epistemic vision. For example, one can see a scarlet diamond without seeing that it

is a scarlet diamond. Seeing that the displayed object is a scarlet diamond is epistemic

seeing. Seeing that an object is a scarlet diamond is coming to believe by visual means

that a perceptually salient object is a scarlet diamond. One cannot come to have this

belief—hence see that an object is a scarlet diamond—unless one has the concepts

scarlet and diamond. There is, however, something it is like to see a scarlet diamond

that is different from seeing a red circle or a green lettuce, even for someone who lacks

both the concept of a diamond and the concept of scarlet. Seeing a scarlet diamond

without seeing that it is a scarlet diamond is non-epistemic seeing.

An interesting case

of weak epistemic seeing is seeing an object as a scarlet diamond.

Introduction

xvii

It is important to emphasize that the distinction between epistemic and non-epistemic seeing does

not entail that only a creature who lacks the concept of a diamond and the concept of scarlet can have non-

epistemic vision of a scarlet diamond. It merely claims that possession of these concepts is not required for

non-epistemic vision.

See Chapter 5 for discussion.

Jacob-FM.qxd 8/23/03 4:55 PM Page xvii

By further distinguishing between primary and secondary epistemic seeing, we try

to analyze the subtle ways in which an individual’s knowledge arises from visual per-

ception and how visual knowledge interacts with non-visual knowledge. In Chapter 6,

we argue that the key towards understanding the modularity of visuomotor representa-

tions is that in a visuomotor representation of a target for action, the visual information

about the shape, size and orientation of the target is trapped in a representation of its

location coded in an egocentric frame of reference. Only by recoding the location of an

object in allocentric coordinates can the visual information about the shape, size and

orientation of an object be available for perceptual processing.

In the early 1980s, electrophysiological and behavioral evidence from macaque

monkeys demonstrated the existence of two separate pathways in the primate visual

system: the ventral pathway and the dorsal pathway. Then in the early 1990s, the

in-depth neuropsychological study of brain-lesioned human patients provided evidence

for the view that the former underlies visual perception, whereas the latter underlies

the visuomotor transformation. We depart from the earlier model of the distinction

between vision-for-perception and vision-for-action in several respects. First of all, as we

have already said, humans can see a great variety of things (including, rivers, substances,

clouds, vapors, smoke, movies, events and actions), only a few of which they can

also grasp and manipulate. On our view, the dualistic model of human vision, accord-

ing to which one and the same object can be visually processed either for the purpose

of visual perception or for the purpose of visual action, primarily applies to the vision

of objects that can be perceived, reached and grasped. As it turns out, seeing human

actions raises issues that cannot be properly understood on the basis of the restricted

duality between the visual perception of objects and the visuomotor trans-

formation. Second, we emphasize the fact that, in the visual life of normally sighted

human adults, perceptual and motor processing of visual inputs do almost always

collaborate. Thirdy, we think that the anatomical distinction between the ventral path-

way and the dorsal pathway, which, according to the earlier dualistic model, underlies

vision-for-perception and vision-for-action, must accommodate the distinction between

the visuomotor transformation and the perceptual processing of spatial properties and

relations among objects. In particular, it must accommodate the dual role of the parietal

lobe, which is both involved in the visuomotor transformation and in the perception of

spatial relationships. Finally, in our view, the visuomotor transformation is but a first

lower level component of the human ‘pragmatic processing’ of objects. We contrast this

lower level pragmatic processing with a higher level pragmatic processing of objects

involved in the skilled use and manipulation of complex cultural tools and artifacts.

Thus, in our view, there is a parallelism between levels of semantic processing and

levels of pragmatic processing. Lower level visuomotor processing stands to higher

level pragmatic processing of objects somewhat as non-epistemic seeing stands to

epistemic seeing on the perceptual side. This parallelism is corroborated by the

neuropsychological evidence. The term ‘agnosia’ was coined by Sigmund Freud to refer

to a visual perceptual impairment. Neuropsychologists make a distinction between

‘apperceptive’ agnosic patients and ‘associative’ agnosic patients. Apperceptive agnosic

patients have a deeper perceptual impairment than associative agnosic patients: the

former fail to process the elementary visual attributes of objects. Although the latter can

process most of the elementary visual attributes of objects (such as their size, shape and

xviii

Introduction

Jacob-FM.qxd 8/23/03 4:55 PM Page xviii

剩余307页未读，继续阅读

top_geek_001

粉丝: 197
资源: 8

视觉认知的边界：《观察方式》解析计算机视觉目标感知

信息安全_数据安全_hum-r1-2-seeing-is-believing-mak.pdf

Drawing-Moving-and-Seeing-with-Code

seeing-theory_ProbabilityTheory_ProbabilityforAI_Seeingtheory_

用英语说说为什么选择从事数据工程师，工作中最美好的部分是什么，最糟糕的部分是什么

could you write a group album which bases on the wechat progromme

onblur.html:20 Uncaught TypeError: Cannot set properties of null

[Error] 'exit' was not declared in this scope

The package-lock.json file was created with an old version of npm,

error in evaluating the argument 'x' in selecting a method for function 'as.matrix'

最新资源