DETAILED ACTION
Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
This communication is responsive to the correspondence filled on 07/14/2025.
Claims 1-20 are presented for examination.
IDS Considerations
The information disclosure statement (IDS) submitted on 07/14/2025 is/are being considered by the examiner as the submission is in compliance with the provisions of 37 CFR 1.97.
Claim Rejections - 35 USC § 102
The following is a quotation of 35 U.S.C. 102(a)(1)/(a)(2) which forms the basis for all obviousness rejections set forth in this Office action:
(a)(1) the claimed invention was patented, described in a printed publication, or in public use, on sale or otherwise available to the public before the effective filing date of the claimed invention.
(a)(2) the claimed invention was described in a patent issued under section 151, or in an application for patent published or deemed published under section 122(b), in which the patent or application, as the case may be, names another inventor and was effectively filed before the effective filing date of the claimed invention.
Claims 20 is/are rejected under 35 U.S.C. 102 (a)(1) as being unpatentable over
Korodi (U.S. Pub. No. 20110248872 A1).
Regarding to claim 20:
Korodi teaches a non-transitory computer readable recording medium storing a bitstream of video which is generated by a method performed by a video processing apparatus (paragraph [0003], [0084], a non-transitory decoder readable storage medium for storing a bitstream generated by an image encoding method or apparatus), wherein the method comprises (the method steps do not carry patentable weight as the claim is a product-by-process claim in which only the bitstream (product), generated by the method steps (process), is given weight. MPEP §2113 recites "Product-by-Process claims are not limited to the manipulations of the recited steps, only the structure implied by the steps". Thus, the scope of the claim is the storage medium storing the bitstream (with the structure implied by the method steps). The structure includes the information and samples manipulated by the steps.
"To be given patentable weight, the printed matter and associated product must
be in a functional relationship. A functional relationship can be found where the printed
matter performs some function with respect to the product to which it is associated".
MPEP §2111.05(I)(A). When a claimed computer-readable medium merely serves as a support for information or data, no functional relationship exists, MPEP §2111.05(III).
The storage medium storing the claimed bitstream in claims 20 merely services as a support for the storage of the bitstream and provides no fictional relationship between the stored bitstream and storage medium. Therefor the structure, which scope is implied by the method steps, is non-functional descriptive material and given no patentable weight. MPEP §2111.05(III). Thus, the claim scope is just a storage medium storing data and is anticipated by Korodi which recites a storage medium storing a bitstream ([0003], [0084].
Any argument attempting to distinguish claim 20 from Korodi will be held not responsive to this claim rejection made on distinct and independent grounds. 37 C.F.R. § 1.111(b).
Claim 20 is directed to a computer readable recording medium [1] storing encoding data generated by a certain image encoding method. This medium storing non-executable content data as digital ones and zeroes is analogous to a printed book in which audio data comprising speech is stored as words written in the Latin alphabet or another writing system. As such, under the “Printed Matter” or “Nonfunctional Descriptive Material” doctrine, the content of the bitstream is not given patentable weight, and the claim is deemed to read “A computer readable medium storing encoding data which is generated by a method of encoding an image”. See M.P.E.P. § 2111.05(III), “When the programming performs some function with respect to the computer with which it is associated, a functional relationship will be found. However, where . . . the computer-readable medium merely serves as a support for information or data, no functional relationship exists”. See also O’Reilly v. Morse, 56 U.S. (15 How.) 62 (1854) (“use of . . . electromagnetism for making or printing intelligible characters, signs, or letters” held to be non-patentable); Ex parte Mathias, 84 U.S.P.Q.2d 1276, 1278–79 (B.P.A.I. 10 August 2005) (informative) (“a computer-readable storage medium that differs from the prior art solely with respect to nonfunctional descriptive material, such as music or a literary work, encoded on the medium” is not distinct from the prior art). Considering this, any computer readable recording medium storing encoding data which is generated by a method of encoding an image, in use or for sale before the effective filing date of 4 October 2016, is considered prior art, for example, a medium storing a video encoded with the H.265 codec.
Claims 1, 2, 4, 13 and 18-20 is/are rejected under 35 U.S.C. 102 (a)(1) as being unpatentable over Hoang Le (MobileCodec: Neural Inter-frame Video Compression on Mobile Devices – MMSys ’22, June 14–17, 2022, Athlone, Ireland - ACM ISBN 978-1-4503-9283-9/22/06. https://doi.org/10.1145/3524273.3532906).
PNG
media_image1.png
200
400
media_image1.png
Greyscale
Regarding to claim 1 and 18-20:
1. Hoang Le teach a method for visual data processing, comprising: (Hoang Le page 1, col 2 para 1 INTRODUCTION - Video compression technologies have been actively researched and engineered over decades to obtain broad video adoption across a wide range of devices, distribution media, and services) determining, for a conversion between a current visual unit of visual data and a bitstream of the visual data, (Hoang Le page 2, col 1 para 2: The main contributions of this paper are: • We present a demonstration of the first real-time neural inter-frame video decoder on a mobile device. • We present an efficient network architecture for a neural inter-frame codec specifically designed for deployment on a mobile platform. • We introduce a parallel entropy coding algorithm tailored for NN-based entropy estimation models) a plurality of threads for coding residual information of the current visual unit; (Hoang Le Fig. 4 page 5 col. 1 para 3: The coding method is an implementation of static (instead of adaptive) arithmetic coding using 32-bit registers, with 16-bit values for probabilities and range, and byte-based renormalization [35]. Since it uses only simple arithmetic and logic operations, it can be executed with multiple threads, without specialized hardware.) and performing the conversion based on the plurality of threads. (Hoang Le Fig. 4 page 4 col 2 last para – page 5 col 1 para 1-2: 3.3 Parallel Entropy Coding - Entropy coding is employed to losslessly compress the quantized latents and hyperlatents to further reduce the rate. This is the stage where the bitstream representing the video is created (at the transmitter) or parsed (at the receiver). For high resolution and quality levels, the entropy coding must support very high throughputs, which can be difficult and computationally expensive without parallelization, even with custom hardware. Although the probability models in our MobileCodec are learned, the probability of each element in 𝑦 and 𝑧 is (conditionally) independent and thus can be coded in parallel. This was implemented by partitioning and indexing the bitstream, providing entry points for parallel decoding [36]. Figure 3 shows an illustration of this process. Concurrent encoders save data to temporary memory buffers, and after frame encoding is completed, those bitstreams are concatenated to create the combined bitstream for that frame. Note that this bitstream must include a header indicating starting positions for each independent part, which are used as entry points for parallel decoding. Hoang Le Fig. 4 page 6 col. 2 last para: We also present an approach to effectively deploy this video codec on a mobile device, including a quantization aware training step to quantize the model using 8-bit integer and a parallel entropy coding algorithm to effectively code the bitstream)
Regarding to claim 2:
2. Hoang Le teach the method of claim 1, wherein a plurality of codewords in the bitstream comprises a plurality of symbols by arithmetic coding in parallel, the residual information being included in the plurality of codewords. (Hoang Le page 5 col. 1 para 3-4: The coding method is an implementation of static (instead of adaptive) arithmetic coding using 32-bit registers, with 16-bit values for probabilities and range, and byte-based renormalization [35]. Since it uses only simple arithmetic and logic operations, it can be executed with multiple threads, without specialized hardware. The integer cumulative distribution arrays needed for arithmetic coding [35] are pre-computed, according to the quantized entropy coding parameter defined by eq. (9), assuming that the latent variables to be encoded have normal probability distributions and are converted to integers using unit-step uniform quantization (i.e., simple rounding). Those arrays are stored at the transmitter and receiver.)
Regarding to claim 4:
4. Hoang Le teach the method of claim 1, wherein the conversion comprises encoding the current visual unit into the bitstream. (Hoang Le Fig. 4)
Regarding to claim 13:
13. Hoang Le teach the method of claim 1, wherein the conversion comprises decoding the current visual unit from the bitstream. (Hoang Le Fig. 4)
Claim Rejections - 35 USC § 103
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
Claims 3, 5 and 14 is/are rejected under 35 U.S.C. 103 as being unpatentable over Hoang Le (MobileCodec: Neural Inter-frame Video Compression on Mobile Devices – MMSys ’22, June 14–17, 2022, Athlone, Ireland - ACM ISBN 978-1-4503-9283-9/22/06. https://doi.org/10.1145/3524273.3532906), in view of Korodi (U.S. Pub. No. 20110248872 A1).
Regarding to claim 3:
3. Hoang Le teach the method of claim 2, Hoang Le do not explicitly teach wherein location information of the plurality of codewords is included in the bitstream, wherein the location information is included in at least one byte at a beginning of the bitstream, wherein the location information comprises a relative offset of a codeword from an initial position in the bitstream, a starting position of the codeword is determined based on the relative offset, or wherein the location information comprises a size of a codeword, a position of the codeword being determined based on the size.
However Korodi teach wherein location information of the plurality of codewords is included in the bitstream, wherein the location information is included in at least one byte at a beginning of the bitstream, wherein the location information comprises a relative offset of a codeword from an initial position in the bitstream, a starting position of the codeword is determined based on the relative offset, or wherein the location information comprises a size of a codeword, a position of the codeword being determined based on the size. (Korodi [0118] FIG. 13, which shows a portion of an encoded subsequence 700. The encoded subsequence is encoded using an arbitrary variable length encoding scheme, such that the resulting encoded subsequence is a series of codewords, shown individually as 702, 704, 706, 708, and 710. The decoding of the previous codeword is not necessary in order to decode any one of the codewords. However, the division of the payload into segments may result in splitting of a codeword, as illustrated by dividing line 720 splitting codeword 706. The dividing line 720 indicates the splitting point determined using an applicable distributor function for defining segment lengths. [0119] In this example embodiment, the decoder requires knowledge of a nearby codeword boundary, such as the beginning of codeword 706 or the end of codeword 706. Accordingly, in one embodiment, the auxiliary information may be an offset defining the number of bits the division or splitting point should be moved in order to align with a codeword boundary. Because the encoder knows the location of the codeword boundaries, it can apply the same distributor function as will be used by the decoder to determine where the decoder will attempt to divide the payload into segments. It may then determine the appropriate offset to align the division with a codeword boundary and inserts the offset in the second field as auxiliary information. [0120] The offset may, by convention, always indicate an offset to either the left or the right. In another embodiment, the offset may be signed to indicate whether the offset is to the left of the right. The offset field may be a predefined length, such as 4, 6 or 8 bits, the length of which will depend on the size of the codewords in the codeset)
It would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to modify Hoang Le, further incorporating Korodi in video/camera technology. One would be motivated to do so, to incorporate location information of the plurality of codewords is included in the bitstream, wherein the location information is included in at least one byte at a beginning of the bitstream, wherein the location information comprises a relative offset of a codeword from an initial position in the bitstream, a starting position of the codeword is determined based on the relative offset. This functionality will improve efficiency with predictable results.
Regarding to claim 5:
5. Hoang Le teach the method of claim 4, Hoang Le do not explicitly teach wherein the number of the plurality of threads comprises a maximum number of threads used in entropy coding.
However Korodi teach wherein the number of the plurality of threads (Korodi [0059] The encoder 200 includes N parallel entropy coders 206 (individually labeled as 206-1, 206-2, . . . , 206-N). Each entropy coder 206 encodes one of the subsequences to produce an encoded subsequence bitstream 210 (individually labeled as 210-1, 210-2, . . . , 210-N). The encoded subsequence bitstreams 210 are then combined into a single bitstream 208 using, for example, a multiplexer 207. In this example, the encoded subsequence bitstreams 210 are multiplexed together to create the bitstream 208 by concatenating the subsequence bitstreams 210 and adding indexing information to the bitstream 208 to enable the decoder to identify the start of each encoded subsequence bitstream 210 in the single bitstream 208.) comprises a maximum number of threads used in entropy coding. (Korodi [0137] In general it would be desirable to set b=d. In general it is also undesirable to set b<d, since one or more of the parallel entropy decoders may sit idle during the decoding process. Accordingly, in accordance with one aspect of the present application, the encoder selects b such that it is likely to be greater than d, but not so large as to impose a heavy bit rate overhead on the bitstream by requiring too much auxiliary information. If b>d then the decoder can allocate the b segments amongst the d decoders so as to best balance the load. [0138] In general it would also be desirable to have b/d be an integer, since the b segments may then be equally allocated amongst the d parallel entropy decoders. [0139] In many cases, a parallel architecture tends to be established using powers of two. In other words, a decoder having parallel entropy decoders is likely to have 2, 4, 8, 16, etc. parallel entropy decoders. Therefore, to take advantage of this likelihood and to minimize the chance that of a unequal allocation of segments to decoders, the number of segments b could be set to a power of 2 sufficiently large to accommodate most architectures, but sufficiently small to avoid excessive overhead in the auxiliary information field. In other words, in some embodiments, the number of segments, b, is set to 2.sup.t. Korodi [0093] In yet another scenario, the encoder does not know in advance how many parallel decoders 306 are available in the decoder. In this case, if the decoder has fewer parallel decoders 306 than the subsequences generated by the encoder, then the decoder cannot process all the subsequences in parallel and the decoder may schedule use of the decoders 306 amongst the subsequences. [0094] For those instances where the number d of parallel decoding processing units differ from N, it would be advantageous to have a mechanism for combining source outputs (for d<N), or splitting source outputs (for d>N). For this example mechanism, the encoder knows the value d at the time of encoding. Nevertheless, if the decoder does not have d decoding units, lossless decoding is still achievable)
Regarding to claim 14:
14. Hoang Le teach the method of claim 13, Hoang Le do not explicitly teach wherein determining the plurality of threads comprises: determining a maximum number of the plurality of threads from the bitstream; and determining a target number of the plurality of threads based on the maximum number and a condition of a device for the conversion, the conversion being performed based on the target number of the plurality of threads.
However Korodi teach wherein determining the plurality of threads comprises: determining a maximum number of the plurality of threads from the bitstream; and determining a target number of the plurality of threads based on the maximum number and a condition of a device for the conversion, the conversion being performed based on the target number of the plurality of threads. (Korodi [0093] In yet another scenario, the encoder does not know in advance how many parallel decoders 306 are available in the decoder. In this case, if the decoder has fewer parallel decoders 306 than the subsequences generated by the encoder, then the decoder cannot process all the subsequences in parallel and the decoder may schedule use of the decoders 306 amongst the subsequences. [0094] For those instances where the number d of parallel decoding processing units differ from N, it would be advantageous to have a mechanism for combining source outputs (for d<N), or splitting source outputs (for d>N). For this example mechanism, the encoder knows the value d at the time of encoding. Nevertheless, if the decoder does not have d decoding units, lossless decoding is still achievable)
Claims 6 is/are rejected under 35 U.S.C. 103 as being unpatentable over Hoang Le (MobileCodec: Neural Inter-frame Video Compression on Mobile Devices – MMSys ’22, June 14–17, 2022, Athlone, Ireland - ACM ISBN 978-1-4503-9283-9/22/06. https://doi.org/10.1145/3524273.3532906), in view of Korodi (U.S. Pub. No. 20110248872 A1), further in view of Huang (U.S. Pub. No. 20100135416 A1 A1).
Regarding to claim 6:
6. Hoang Le teach the method of claim 5, Hoang Le do not explicitly teach wherein the maximum number of threads is determined based on at least one of: a profile, or a configuration file, or wherein the maximum number of threads is a fixed parameter, or wherein the maximum number of threads is determined based on at least one of: the number of slices, the number of tiles, the number of processing units, or a resolution of a processing unit of the visual data.
However Huang teach wherein the maximum number of threads is determined based on at least one of: a profile, or a configuration file, or wherein the maximum number of threads is a fixed parameter, or wherein the maximum number of threads is determined based on at least one of: the number of slices, the number of tiles, the number of processing units, or a resolution of a processing unit of the visual data. (Huang Fig. 1-2 [0027] In Step 920, under control of the controller 114, the processing elements 112-1, 112-2, . . . , and 1112-Z [threads] respectively start to perform a coding procedure for the plurality of entropy slices [processing units] according to the causal criterion, so that a number of the entropy slices are processed in parallel. Fig. 5-8 [0045] FIG. 7 illustrates an exemplary timing diagram of parallel video decoding according to another embodiment, where 16 processing elements (e.g. PE00, PE01, PE02, PE03, . . . , and PE15) are utilized for decoding 64 entropy slices (e.g. the entropy slices ES00, ES01 , ES02, ES03, . . . , and ES63). Each processing element is responsible for a number of entropy slices, and it is not limited to assign the same amount of entropy slices to each processing element as shown in FIG. 7. The predetermined delay amount corresponding to the processing time of D.sub.A MBs is applied to the control of the start time points. For example, before the apparatus 100 starts to parse a current entropy slice (e.g. entropy slice ES03), D.sub.A MBs of the previous entropy slice thereof (e.g. entropy slice ES02) have been parsed.)
The motivation for combining Hoang Le and Korodi as set forth in claim 3 is equally applicable to claim 6. It would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to modify Hoang Le, further incorporating Korodi and Huang in video/camera technology. One would be motivated to do so, to incorporate the maximum number of threads is determined based on at least one of: a profile, or a configuration file, or wherein the maximum number of threads is a fixed parameter, or wherein the maximum number of threads is determined based on at least one of: the number of slices, the number of tiles, the number of processing units, or a resolution of a processing unit of the visual data. This functionality will improve quality with predictable results.
5. Claims 17 is/are rejected under 35 U.S.C. 103 as being unpatentable over Hoang Le (MobileCodec: Neural Inter-frame Video Compression on Mobile Devices – MMSys ’22, June 14–17, 2022, Athlone, Ireland - ACM ISBN 978-1-4503-9283-9/22/06. https://doi.org/10.1145/3524273.3532906), in view of Estevez (U.S. Pub. No. 20080085031 A1).
Regarding to claim 17:
17. Hoang Le teach the method of claim 1, wherein a byte alignment is performed at an end of a codeword in the bitstream, wherein a bit length of the codeword is not a multiple of a predefined bit length, the byte alignment is performed at the end of the codeword by adding stuffing bits, a bit length of the aligned codeword being of a multiple of the predefined bit length, wherein the predefined bit length is 8 bits, and/or (Part of OR condition, no rejection is required)
Hoang Le do not explicitly teach wherein the byte alignment is applied to a plurality of codewords in the bitstream.
However Estevez teach wherein the byte alignment is applied to a plurality of codewords in the bitstream. (Estevez [0020] For a standard such as H.263, DC coefficients of intra-coded macroblocks are fixed-length coded with 8 bits. For a standard such as MPEG-4 without short headers, the value of DC coefficients is variable. In either case, bits of data can be encoded by altering the DC coefficients. If necessary, a different size codeword could be used. In such case, the byte alignment or header information regarding a coded frame size can be adjusted accordingly. In at least some embodiments, the byte alignment can be corrected by adjusting the bit stuffing used to align start codes. As needed, extra bits (4 bytes or 8 bytes) could be inserted per renderer for the network address or device identifier. In summary, for some embodiments, watermarking videos may be accomplished by adjusting DO or AC coefficients. Alternative or similar watermarking techniques could be used for audio, video games, photos or slide shows. [0021] In at least some embodiments, adding a watermark requires a full transcoding of the bitstream. Alternatively, a simplified (low complexity) watermarking technique can be implemented by modifying fixed-length codewords that do not alter the processing flow of the decoder. Also, variable-length codewords may be modified. If the length of a codeword is altered, the byte alignment of certain start codes and the overall bit rate are maintained. To reduce the visibility of a watermark, some embodiments perform watermarking with only minor changes being made to any single codeword.)
It would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to modify Hoang Le, further incorporating Estevez in video/camera technology. One would be motivated to do so, to incorporate the byte alignment is applied to a plurality of codewords in the bitstream. This functionality will improve accuracy with predictable results.
Allowable subject matter
Regarding to claim 7-12 and 15-16:
Claims 7-12 and 15-16 is/are objected to as being dependent upon a rejected base claim, but would be allowable if rewritten in independent form including all of the limitations of the base claim and any intervening claims because the limitations of these dependent claims are not obvious from the prior art search when all the limitations of independent and intervening claims are taken into account.
Regarding to claim 7:
7. Hoang Le teach the method of claim 5, wherein performing the conversion comprises: determining at least one quantized residual representation and at least one corresponding quantized hyperprior information of the current visual unit; (Hoang Le page 2, col 2 para 3: Intra-frame MobileCodec. The design of our Intra-frame Mobile-Codec is derived from the image compression method by Balle etal. [5]. Please refer to the original paper for a detailed description of the method. Briefly, as shown in Figure 1, the encoder (or analysis transform) 𝐼𝑎 transforms each frame 𝑥0 into its embedded latent 𝑦𝐼 = 𝐼𝑎 (𝑥0). This embedded latent 𝑦𝐼 is then quantized by the rounding operator 𝑄, and entropy coded by an arithmetic encoder (𝐴𝐸) before being sent to the receiver where it will be decoded by an arithmetic decoder (𝐴𝐷) and reconstructed back to the image ˆ 𝑥0 via the synthesis decoder 𝐼𝑠 : ˆ 𝑥0 = 𝐼𝑠 (𝑦𝐼 ). Following [5], we also use a hypercodec network to model the density of the encoded latent 𝑦𝐼 . Specifically, the hyper-encoder 𝐻𝐼 𝑎 transforms the latent 𝑦𝐼 into 𝑧𝐼 = 𝐻𝐼 𝑎 (𝑦𝐼 ). 𝑧𝐼 is also quantized, entropy coded, and sent to the receiver where it will be decoded and fed into a hyper-decoder network 𝐻𝐼 𝑠 to predict the scale 𝜎𝐼 of the latent 𝑦ˆ𝐼 : 𝜎𝐼 = 𝐻𝐼 𝑠 ( ˆ 𝑧𝐼 ). The predicted scale 𝜎𝐼 is used for arithmetic encoding 𝐴𝐶 only in the transmitter side and arithmetic decoding 𝐴𝐷 in both transmitter and receiver sides for losslessly coding the quantized latent 𝑦ˆ𝐼 . Note that, in contrast to [5], we as use ReLU as the nonlinearity after each layer of the network instead of a GDN block. Our study showed that using ReLU is more quantization friendly for MobileCodec while still being able to obtain comparable performance.)
Prior art do not teach determining a first probability distribution of the at least one quantized residual representation and a second probability distribution of the at least one quantized hyperprior information; dividing the at least one quantized residual representation into a plurality of groups of quantized residual representations based on the number of the plurality of threads; determining a plurality of codewords based on the plurality of groups of quantized residual representations; and generating the bitstream based on the plurality of codewords.
Regarding to claim 15:
15. Hoang Le teach the method of claim 14, wherein performing the conversion comprises: reading a plurality of locations in the bitstream based on the plurality of threads; determining a probability of quantized hyperprior information; decoding the quantized hyperprior information by arithmetic coding; (Hoang Le page 2, col 2 para 3: Intra-frame MobileCodec. The design of our Intra-frame Mobile-Codec is derived from the image compression method by Balle etal. [5]. Please refer to the original paper for a detailed description of the method. Briefly, as shown in Figure 1, the encoder (or analysis transform) 𝐼𝑎 transforms each frame 𝑥0 into its embedded latent 𝑦𝐼 = 𝐼𝑎 (𝑥0). This embedded latent 𝑦𝐼 is then quantized by the rounding operator 𝑄, and entropy coded by an arithmetic encoder (𝐴𝐸) before being sent to the receiver where it will be decoded by an arithmetic decoder (𝐴𝐷) and reconstructed back to the image ˆ 𝑥0 via the synthesis decoder 𝐼𝑠 : ˆ 𝑥0 = 𝐼𝑠 (𝑦𝐼 ). Following [5], we also use a hypercodec network to model the density of the encoded latent 𝑦𝐼 . Specifically, the hyper-encoder 𝐻𝐼 𝑎 transforms the latent 𝑦𝐼 into 𝑧𝐼 = 𝐻𝐼 𝑎 (𝑦𝐼 ). 𝑧𝐼 is also quantized, entropy coded, and sent to the receiver where it will be decoded and fed into a hyper-decoder network 𝐻𝐼 𝑠 to predict the scale 𝜎𝐼 of the latent 𝑦ˆ𝐼 : 𝜎𝐼 = 𝐻𝐼 𝑠 ( ˆ 𝑧𝐼 ). The predicted scale 𝜎𝐼 is used for arithmetic encoding 𝐴𝐶 only in the transmitter side and arithmetic decoding 𝐴𝐷 in both transmitter and receiver sides for losslessly coding the quantized latent 𝑦ˆ𝐼 . Note that, in contrast to [5], we as use ReLU as the nonlinearity after each layer of the network instead of a GDN block. Our study showed that using ReLU is more quantization friendly for MobileCodec while still being able to obtain comparable performance.)
Prior art do not teach determining a probability of at least one quantized residual representation based on the probability of quantized hyperprior information; decoding the at least one quantized residual representation based on the plurality of codewords in parallel; and determining a reconstructed image based on the decoded quantized residual representation.
Closely related prior art
Examiner notes teaching of U.S. Pub. No. 20240380929 A1 is/are pertinent to the independent claim(s).
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to NASIM N NIRJHAR whose telephone number is (571) 272-3792. The examiner can normally be reached on Monday - Friday, 8 am to 5 pm ET.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, William F Kraig can be reached on (571) 272-8660. The fax phone number for the organization where this application or proceeding is assigned is (571) 273-8300.
Information regarding the status of an application may be obtained from the Patent Application Information Retrieval (PAIR) system. Status information for published applications may be obtained from either Private PAIR or Public PAIR. Status information for unpublished applications is available through Private PAIR only. For more information about the PAIR system, see http://pair-direct.uspto.gov. Should you have questions on access to the Private PAIR system, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative or access to the automated information system, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/NASIM N NIRJHAR/Primary Examiner, Art Unit 2896
Prosecution Insights