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 .
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 1, 8, 15 are rejected under 35 U.S.C. 103 as being unpatentable over Chen et al. (US20250247586A1), hereinafter Chen, in view of Hammoud et al.: "SynthCLIP: Are We Ready for a Fully Synthetic CLIP Training?", Arxiv, Published 18 Jul 2024 [Retrieved on 09/30/2026]. Retrieved from google scholar <https://arxiv.org/abs/2402.01832>, hereinafter Hammoud, and further in view of Wang et al.: "ModelScope Text-to-Video Technical Report", Arxiv, Published 12 Aug 2023 [Retrieved on 09/30/2026]. Retrieved from google scholar <https://arxiv.org/abs/2308.06571>, hereinafter Wang.
Regarding claim 1, Chen teaches a method comprising: receiving a video as an input to a video captioning model (Para. 13 see "A large number of high-resolution videos may be curated from publicly available datasets, where the videos are split into semantically consistent video clips, and then multiple cross-modality teacher models are applied to obtain captions for each video clip."; Para. 14 see "In examples described herein, 70 million (70M) videos pairs are generated with high-quality text captions, referred to generally in this disclosure as a Panda-70M dataset."; Para. 21 see "However, by jointly collecting all the captions from different models, it is observed that 88.8% of videos can be annotated with at least one good caption."; Para. 22 see "a range of cross-modality teacher models 208 are used, including image captioning models and image/video visual-question-answering (VQA) models with an additional text input 210, such as video description and subtitles, to predict several candidate captions for a video clip."; Para. 42 see "For the vision branch 402, the Video-LLaMA is used to extract LLM-compatible video embedding."; Para. 45 see "To quantitatively evaluate the quality of output captions, common protocols are followed and report BLEU-4, ROGUE-L, METEOR, and CIDEr."; Para. 55 see "FIG. 8 is a diagrammatic representation of the machine 800 within which instructions 810 (e.g., software, a program, an application, an applet, an app, or other executable code) for causing the machine 800 to perform any one or more of the methodologies discussed herein may be executed. For example, instructions 810 may cause the machine 800 to execute any one or more of the methods described herein including for automatic captioning pipeline 200. ... The machine 800 may operate as a standalone device or may be coupled (e.g., networked) to other machines."; Para. 64 see "The various memories (e.g., main memory 814, static memory 816, and memory of the processors 804) and storage unit 818 may store one or more sets of instructions and data structures (e.g., software) embodying or used by any one or more of the methodologies or functions described herein. These instructions (e.g., the instructions 810), when executed by processors 804, cause various operations to implement the disclosed examples including automatic captioning pipeline 200.") generating one or more captions using the video captioning model (Para. 13 see "A large number of high-resolution videos may be curated from publicly available datasets, where the videos are split into semantically consistent video clips, and then multiple cross-modality teacher models are applied to obtain captions for each video clip."; Para. 22 see "Second, a range of cross-modality teacher models 208 are used, including image captioning models and image/video visual-question-answering (VQA) models with an additional text input 210, such as video description and subtitles, to predict several candidate captions for a video clip."; Para. 32 see "The automatic captioning pipeline 200 having multiple cross-modality teacher models 208 is used. In an example, a large pool including 31 captioning teacher models 208 is used.") outputting the one or more captions (Para. 14 see "In examples described herein, 70 million (70M) videos pairs are generated with high-quality text captions, referred to generally in this disclosure as a Panda-70M dataset."; Para. 21 see "However, by jointly collecting all the captions from different models, it is observed that 88.8% of videos can be annotated with at least one good caption."; Para. 38 see "Given multiple candidate captions for a video, the caption that best aligns to the video content is used."; Para. 45 see "To quantitatively evaluate the quality of output captions, common protocols are followed and report BLEU-4, ROGUE-L, METEOR, and CIDEr.") wherein the video captioning model is trained using a large language model that receives a prompt (Para. 23 see "To achieve efficient video annotation at scale, in one example, the system uses a student captioning model that is trained to distill the knowledge from the teachers models 208."; Para. 32 see "for the VQA models, in addition to visual data, a prompt is also input, including additional text information, such as video description and subtitles, and the teacher models 208 are asked to summarize all multimodal inputs into one sentence."; Para. 34 see "Video-LLaMA: For all teacher models 208, the system uses the vision branch and not the audio branch and uses Vicuna-7B as the LLM."; Para. 42 see "For the text branch 404, text embedding extracted by a text encoder 406 is directly input into the LLM 408. ... example systems include a text Q-former 410 that extracts the text embedding with fixed length and better bridges the video and text embeddings.") finetuning the video captioning model using video-caption pairs (Para. 23 see "To achieve efficient video annotation at scale, in one example, the system uses a student captioning model that is trained to distill the knowledge from the teachers models 208"; Para. 41 see "According to this disclosure, a student captioning model is trained on the Panda-70M dataset to distill the knowledge from multiple teacher models 208."; Para. 44 see "The student captioning model 400 is also trained with both video and text branches 402 and 404, respectively, on the complete Panda-70M dataset for better captioning performance.").
While Chen teaches using LLM teachers that prompt on title, description, and subtitles to caption real videos, then training a student captioner on those pairs, Chen does not teach a large language model that receives a prompt related to a content category generates a plurality of captions based on the prompt wherein the plurality of captions is provided to a text-to-video model that generates video counterparts each associated with a caption of the plurality of captions the pairs include the plurality of captions generated by the large language model each pair includes a video counterpart generated from an associated caption.
However, Hammoud teaches a large language model that receives a prompt related to a content category (Pg. 3, Para. 2 see "The first stage of our pipeline involves the generation of synthetic image captions, that we later aim to use as prompts for text-to-image generators. To achieve this, we utilize an LLM conditioned on our concept bank. The model is prompted to generate captions that describe a scene related to a chosen concept."; Pg. 3, Para. 3 see "We start by introducing our concept bank C composed by NC concepts. Unless otherwise stated, we use the MetaCLIP concept bank (Xu et al., 2023), that contains over 500,000 concepts drawn from WordNet Synsets and Wikipedia common unigrams, bigrams, and titles."; Pg. 4, Para. 1 see "Assuming c ∈ C, our designed prompt is: Your task is to write me an image caption that includes and visually describes a scene around a concept. Your concept is c. Output one single grammatically correct caption that is no longer than 15 words. Do not output any notes, word counts, facts, etc. Output one single sentence only.") generates a plurality of captions based on the prompt (Abstract see "Leveraging recent text-to-image (TTI) networks and large language models (LLM), we generate synthetic datasets of images and corresponding captions at scale, with no human intervention."; Pg. 3, Fig. 1 see "From a set of concepts C (left), we obtain a set of synthetic captions T with an LLM, further refined to T ∗ by a filtering balanced sampling operation (top)."; Pg. 4, Para. 1 see "Assuming c ∈ C, our designed prompt is: Your task is to write me an image caption that includes and visually describes a scene around a concept. Your concept is c."; Pg. 4, Para. 2 see "Formally, we define our LLM generator as GLLM and the prompt as p. Hence, the set of generated captions is T = {tc,n ∼ GLLM(p, c)}, ∀c ∈ C, ∀n ∈ {1, 2,..., N} where N is the number of desired captions for each concept.") the pairs include the plurality of captions generated by the large language model (Pg. 3, Fig. 1 see "From a set of concepts C (left), we obtain a set of synthetic captions T with an LLM, further refined to T ∗ by a filtering balanced sampling operation (top)."; Pg. 4, Para. 2 see "Formally, we define our LLM generator as GLLM and the prompt as p. Hence, the set of generated captions is T = {tc,n ∼ GLLM(p, c)}, ∀c ∈ C, ∀n ∈ {1, 2,..., N} where N is the number of desired captions for each concept."; Pg. 5, Para. 1 see "Formally, we extract representations h = Eimage(xk), xk ∈ I∗ and z = Etext(tk), tk ∈ T∗, and train by minimizing the CLIP loss LCLIP(h, z).").
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified Chen to incorporate the teachings of Hammoud to use a content category prompt so a language model outputs many captions and include those captions as the text in the pairs used to train the video captioning model. Doing so would predictably improve robustness by covering many topics in the training text instead of only the captions that come with collected videos.
Furthermore, Wang teaches wherein the plurality of captions is provided to a text-to-video model that generates video counterparts each associated with a caption of the plurality of captions (Pg. 2, Fig. 1 see "Figure 1: The qualitative results of our ModelScopeT2V model.The text-to-video-synthesis model could generate video frames adhering to the given prompts."; Pg. 3, Para. 3 see "Given a text prompt p, ModelScopeT2V outputs a video vpr through a latent video diffusion model that conforms to the semantic meaning of the prompt."; Pg. 3, Para. 5 see "ModelScopeT2V aims to generate videos conforming to the given text prompts.") each pair includes a video counterpart generated from an associated caption (Pg. 2, Fig. 1 see "Figure 1: The qualitative results of our ModelScopeT2V model.The text-to-video-synthesis model could generate video frames adhering to the given prompts."; Pg. 3, Para. 3 see "Given a text prompt p, ModelScopeT2V outputs a video vpr through a latent video diffusion model that conforms to the semantic meaning of the prompt."; Pg. 3, Para. 5 see "ModelScopeT2V aims to generate videos conforming to the given text prompts").
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified Chen in view of Hammoud to incorporate the teachings of Wang to feed the captions into a text-to-video model so each caption is paired with a generated video counterpart in the training set. Doing so would predictably improve captioning accuracy by giving the model matching text-video pairs instead of relying only on collected videos.
Claim 8 is rejected under the same analysis as claim 1 above.
Claim 15 is rejected under the same analysis as claim 1 above.
Claims 2-4, 9-11, 16-18 are rejected under 35 U.S.C. 103 as being unpatentable over Chen et al. (US20250247586A1), hereinafter Chen, in view of Hammoud et al.: "SynthCLIP: Are We Ready for a Fully Synthetic CLIP Training?", Arxiv, Published 18 Jul 2024 [Retrieved on 09/30/2026]. Retrieved from google scholar <https://arxiv.org/abs/2402.01832>, hereinafter Hammoud, and further in view of Wang et al.: "ModelScope Text-to-Video Technical Report", Arxiv, Published 12 Aug 2023 [Retrieved on 09/30/2026]. Retrieved from google scholar <https://arxiv.org/abs/2308.06571>, hereinafter Wang and Shvetsova et al.: "HowToCaption: Prompting LLMs to Transform Video Annotations at Scale", Arxiv, Published 7 Sep 2024 [Retrieved on 09/30/2026]. Retrieved from google scholar <https://arxiv.org/abs/2310.04900>, hereinafter Shvetsova.
Regarding claim 2, Chen in view of Hammoud and further in view of Wang teaches the method of claim 1.
In addition, Chen teaches wherein the video captioning model is further trained using video-caption pairs (Para. 23 see "To achieve efficient video annotation at scale, in one example, the system uses a student captioning model that is trained to distill the knowledge from the teachers models 208."; Para. 41 see "According to this disclosure, a student captioning model is trained on the Panda-70M dataset to distill the knowledge from multiple teacher models 208."; Para. 44 see "The student captioning model 400 is also trained with both video and text branches 402 and 404, respectively, on the complete Panda-70M dataset for better captioning performance.").
While Chen teaches picking the caption that best matches each video and training the captioning model on those pairs, Chen does not teach creating video-caption pairs based on the generation of the video counterparts a similarity score for each of the video-caption pairs.
However, Wang teaches creating video-caption pairs based on the generation of the video counterparts (Pg. 2, Fig. 1 see "Figure 1: The qualitative results of our ModelScopeT2V model.The text-to-video-synthesis model could generate video frames adhering to the given prompts."; Pg. 3, Para. 3 see "Given a text prompt p, ModelScopeT2V outputs a video vpr through a latent video diffusion model that conforms to the semantic meaning of the prompt."; Pg. 3, Para. 5 see "ModelScopeT2V aims to generate videos conforming to the given text prompts.").
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified Chen in view of Hammoud and further in view of Wang to incorporate the teachings of Wang to create the video-caption training pairs by generating a matching video for each caption. Doing so would predictably increase training coverage by adding video-caption pairs that do not already have a collected real clip.
Furthermore, Shvetsova teaches a similarity score for each of the video-caption pairs (Pg. 7, Para. 3 see "we use the V-L model to compute alignment similarity scores ρn,i(δ) = sim f (cn,i), g(V [τ s n,i+δ,τ e n,i+δ] n) between the caption and video clips with time offsetsδ ∈ Z, /δ/ ≤T around predicted timestamps."; Pg. 7, Para. 3 see "Then we align the caption to the video clip by finding the best offset around the timestamp δ∗ n,i = arg max δ∈{−T,...,T} ρn,i(δ) and filter out pairs if ρn,i(δ∗ n,i) < κ, where κ is a similarity score threshold."; Pg. 11, Para. 1 see "For HowToCaption method, we use T = 10 seconds offset for alignment and adaptive threshold κ to leave 25M most similar pairs after filtering.").
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified Chen in view of Hammoud and further in view of Wang to incorporate the teachings of Shvetsova to use a similarity score for each video-caption pair when those pairs are used to train the captioning model. Doing so would predictably improve accuracy by keeping pairs whose video and caption actually match and dropping pairs that do not.
Regarding claim 3, Chen in view of Hammoud and further in view of Wang and Shvetsova teaches the method of claim 2.
While Chen teaches picking the caption that best matches each video and training the captioning model on those pairs, Chen does not teach wherein the similarity score for each of the video-caption pairs is determined using a multimodal foundational model.
However, Shvetsova teaches wherein the similarity score for each of the video-caption pairs is determined using a multimodal foundational model (Pg. 7, Para. 3 see "we further improve our obtained captions with an alignment & filtering post-processing step (Fig. 2). To this end, we utilize the video-language encoder model(f, g). ... we use the V-L model to compute alignment similarity scores ρn,i(δ) = sim f (cn,i), g(V [τ s n,i+δ,τ e n,i+δ] n) between the caption and video clips with time offsetsδ ∈ Z, /δ/ ≤T around predicted timestamps.").
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified Chen in view of Hammoud and further in view of Wang and Shvetsova to incorporate the teachings of Shvetsova to determine the similarity score for each video-caption pair with a multimodal foundational model. Doing so would predictably improve pairing accuracy by comparing video and text in one shared space.
Regarding claim 4, Chen in view of Hammoud and further in view of Wang and Shvetsova teaches the method of claim 2.
In addition, Chen teaches wherein: the video captioning model is further trained using one or more selected sets of the video-caption pairs (Para. 13 see "A retrieval model is fine-tuned on a relatively small subset ... then the retrieval model is applied to the dataset in its entirety to select the best caption as the annotation."; Para. 23 see "To achieve efficient video annotation at scale, in one example, the system uses a student captioning model that is trained to distill the knowledge from the teachers models 208."; Para. 41 see "According to this disclosure, a student captioning model is trained on the Panda-70M dataset to distill the knowledge from multiple teacher models 208."; Para. 44 see "The student captioning model 400 is also trained with both video and text branches 402 and 404, respectively, on the complete Panda-70M dataset for better captioning performance.").
While Chen teaches picking the caption that best matches each video and training the captioning model on those pairs, Chen does not teach video-caption pairs selected by filtering the video-caption pairs based on the similarity scores and the finetuning of the video captioning model uses the one or more selected sets of video-caption pairs.
However, Shvetsova teaches video-caption pairs selected by filtering the video-caption pairs based on the similarity scores (Pg. 7, Para. 3 see "we use the V-L model to compute alignment similarity scores ρn,i(δ) = sim f (cn,i), g(V [τ s n,i+δ,τ e n,i+δ] n) between the caption and video clips with time offsetsδ ∈ Z, /δ/ ≤T around predicted timestamps."; Pg. 7, Para. 3 see "Then we align the caption to the video clip by finding the best offset around the timestamp δ∗ n,i = arg max δ∈{−T,...,T} ρn,i(δ) and filter out pairs if ρn,i(δ∗ n,i) < κ, where κ is a similarity score threshold."; Pg. 8, Para. 3 see "After alignment & filtering (details in Sec. 4.2) we obtain 25M high-quality video-caption pairs."; Pg. 11, Para. 1 see "For HowToCaption method, we use T = 10 seconds offset for alignment and adaptive threshold κ to leave 25M most similar pairs after filtering.") and the finetuning of the video captioning model uses the one or more selected sets of video-caption pairs (Abstract see "the resulting captions not only significantly improve the performance over many different benchmark datasets for zero-shot text-video retrieval and video captioning, but also lead to a disentangling of textual narration from the audio, boosting the performance in text-video-audio tasks."; Pg. 8, Para. 1 see "we fine-tune (c.f. Sec. 3.2) the V-L model on the aligned & filtered video-captions pairs{(vi, ci)}").
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified Chen in view of Hammoud and further in view of Wang and Shvetsova to incorporate the teachings of Shvetsova to filter the video-caption pairs using the similarity scores and finetune the captioning model on the retained set. Doing so would predictably improve captioning accuracy by training only on pairs whose video and text actually match.
Claim 9 is rejected under the same analysis as claim 2 above.
Claim 10 is rejected under the same analysis as claim 3 above.
Claim 11 is rejected under the same analysis as claim 4 above.
Claim 16 is rejected under the same analysis as claim 2 above.
Claim 17 is rejected under the same analysis as claim 3 above.
Claim 18 is rejected under the same analysis as claim 4 above.
Claims 7, 14 are rejected under 35 U.S.C. 103 as being unpatentable over Chen et al. (US20250247586A1), hereinafter Chen, in view of Hammoud et al.: "SynthCLIP: Are We Ready for a Fully Synthetic CLIP Training?", Arxiv, Published 18 Jul 2024 [Retrieved on 09/30/2026]. Retrieved from google scholar <https://arxiv.org/abs/2402.01832>, hereinafter Hammoud, and further in view of Wang et al.: "ModelScope Text-to-Video Technical Report", Arxiv, Published 12 Aug 2023 [Retrieved on 09/30/2026]. Retrieved from google scholar <https://arxiv.org/abs/2308.06571>, hereinafter Wang and Polyak et al.: "Movie Gen: A Cast of Media Foundation Models", Arxiv, Published 26 Feb 2025 [Retrieved on 09/30/2026]. Retrieved from google scholar <https://arxiv.org/abs/2410.13720>, hereinafter Polyak.
Regarding claim 7, Chen in view of Hammoud and further in view of Wang teaches the method of claim 1.
While Chen teaches picking the caption that best matches each video and training the captioning model on those pairs, Chen does not teach wherein the video counterparts are generated from a caption a plurality of video counterparts generated from a single caption using different seeds.
However, Wang teaches wherein the video counterparts are generated from a caption (Pg. 2, Fig. 1 see "Figure 1: The qualitative results of our ModelScopeT2V model.The text-to-video-synthesis model could generate video frames adhering to the given prompts."; Pg. 3, Para. 3 see "Given a text prompt p, ModelScopeT2V outputs a video vpr through a latent video diffusion model that conforms to the semantic meaning of the prompt."; Pg. 3, Para. 5 see "ModelScopeT2V aims to generate videos conforming to the given text prompts.").
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified Chen in view of Hammoud and further in view of Wang to incorporate the teachings of Wang to generate each video counterpart from a caption. Doing so would predictably increase training coverage by turning written captions into matching videos instead of relying only on collected real clips.
Furthermore, Polyak teaches a plurality of video counterparts generated from a single caption using different seeds (Pg. 83, Para. 5 see "Specifically, for each prompt, we generate 5 different videos from Movie Gen Video using different random seeds. From each of these 5 videos, we manually pick the “best”.").
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified Chen in view of Hammoud and further in view of Wang to incorporate the teachings of Polyak to generate several video counterparts from one caption by using different seeds. Doing so would predictably improve robustness by giving the captioning model more visual versions of the same text without writing new captions.
Claim 14 is rejected under the same analysis as claim 7 above.
Claims 6, 13, 20 are rejected under 35 U.S.C. 103 as being unpatentable over Chen et al. (US20250247586A1), hereinafter Chen, in view of Hammoud et al.: "SynthCLIP: Are We Ready for a Fully Synthetic CLIP Training?", Arxiv, Published 18 Jul 2024 [Retrieved on 09/30/2026]. Retrieved from google scholar <https://arxiv.org/abs/2402.01832>, hereinafter Hammoud, and further in view of Wang et al.: "ModelScope Text-to-Video Technical Report", Arxiv, Published 12 Aug 2023 [Retrieved on 09/30/2026]. Retrieved from google scholar <https://arxiv.org/abs/2308.06571>, hereinafter Wang, Shvetsova et al.: "HowToCaption: Prompting LLMs to Transform Video Annotations at Scale", Arxiv, Published 7 Sep 2024 [Retrieved on 09/30/2026]. Retrieved from google scholar <https://arxiv.org/abs/2310.04900>, hereinafter Shvetsova, and Wu et al.: "GODIVA: Generating Open-Domain Videos from Natural Descriptions", Arxiv, Published 30 Apr 2021 [Retrieved on 09/30/2026]. Retrieved from google scholar <https://arxiv.org/abs/2104.14806>, hereinafter Wu.
Regarding claim 6, Chen in view of Hammoud and further in view of Wang and Shvetsova teaches the method of claim 2.
In addition, Chen teaches video-caption pairs (Para. 23 see "To achieve efficient video annotation at scale, in one example, the system uses a student captioning model that is trained to distill the knowledge from the teachers models 208."; Para. 41 see "According to this disclosure, a student captioning model is trained on the Panda-70M dataset to distill the knowledge from multiple teacher models 208."; Para. 44 see "The student captioning model 400 is also trained with both video and text branches 402 and 404, respectively, on the complete Panda-70M dataset for better captioning performance.").
While Chen teaches picking the caption that best matches each video and training the captioning model on those pairs, Chen does not teach wherein, to determine the similarity score for each of the video-text pairs: a text embedding is created using a text of one of the video-text pairs; a plurality of frame embeddings is created using the video counterparts associated with the text; a semantic similarity is measured for each of the plurality of frame embeddings and the text embedding; and the semantic similarities are combined into the similarity score.
However, Wu teaches wherein, to determine the similarity score for each of the video-text pairs: a text embedding is created using a text of one of the video-text pairs; a plurality of frame embeddings is created using the video counterparts associated with the text; a semantic similarity is measured for each of the plurality of frame embeddings and the text embedding; and the semantic similarities are combined into the similarity score (Pg. 5, Para. 6 see "The key factor for judging the quality of the generated video is whether it matches the text."; Pg. 5, Para. 6 see "Thanks to recent zero-shot work CLIP [18], which provides a strong zero-shot ability for visual-text matching and thus reduced those data biases. Since CLIP is pretrained between image and text, we calculate the similarities between text and each frame of the video and then take the average value as the semantic matching in Eq. (14)."; Pg. 7, Para. 2 see "Eq. (14) where t denotes the input text. ˆv is the predicted video with L frames. Note that SIM only provides the absolute score of the semantic match.").
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified Chen in view of Hammoud and further in view of Wang and Shvetsova to incorporate the teachings of Wu to form the similarity score by embedding the caption, embedding multiple frames of the paired video, measuring each frame against that caption, and combining those measurements. Doing so would predictably improve score reliability and increase accuracy by using the whole clip instead of a single frame that may misrepresent the overall content of the video.
Claim 13 is rejected under the same analysis as claim 6 above.
Claim 20 is rejected under the same analysis as claim 6 above.
Allowable Subject Matter
Claims 5, 12, 19 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.
Regarding claims 5, 12, 19, none of the references cited above teach wherein, to finetune the video captioning model, contrastive learning is applied in both a caption-wise and a category-wise manner.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure. Pei et al. (US20210281774A1) discloses a video caption generating method.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to ALEXANDER VAUGHN whose telephone number is (571) 272-5253. The examiner can normally be reached M-F 11am-7pm (ET).
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, JENNIFER MEHMOOD can be reached at (571) 272-2976. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and manage patent submissions in Patent Center, visit: https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format. For additional questions, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/ALEXANDER VAUGHN/
Examiner, Art Unit 2675
/ATIBA O FITZPATRICK/Primary Examiner, Art Unit 2677
