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 Interpretation
The claims will be read under the broadest reasonable interpretation standard outlined in
MPEP § 2111.01.
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 1-2, 10-12, and 20 are rejected under 35 U.S.C. 103 as being unpatentable over Zhao et. al (US 20240367685 A1) (Hereinafter, “Zhao”) in view of Yang et. al (US 20240409124 A1) (Hereinafter, “Yang”)
With respect to claim 1, Zhao teaches:
A method of detecting a three-dimensional (3D) object ([Abstract]), the method comprising
receiving an input image with respect to a 3D space, an input point cloud with respect to 3D space ([0031] “In some implementations, the sensor subsystem 104 can classify groups of raw sensor measurements from one or more sensors, e.g., a camera sensor, a LiDAR sensor, or both, as being measures of another agent in the environment”; [0002] “Three-dimensional object detection neural networks receive input sensor data, e.g., point cloud or camera image data”), and an input language with respect to a target object in the 3D space ([0085] “Additionally or alternatively, the REM query system 600 can receive a query that includes text 604 and use the selection engine 620 to retrieve one or more sensor inputs 624 based on the text 604”)
using an encoding model to generate candidate image features of partial areas of the input image ([0047] “For each sensor input 211 included in the dataset 210, the system 200 uses the encoder neural network 220 to generate one or more feature vectors for the sensor input. Then, for each of the one or more feature vectors, the system 200 uses the density estimation model to generate a density score for the feature vector”; [0044] “Each sensor input 211 can include, e.g., a LiDAR point cloud generated by a LiDAR sensor, an image captured by camera sensor, or a fused sensor input that combines data from multiple sensors”) a point cloud feature of the input point cloud ([0044]; [0063] “In the example of FIG. 4 , the rare example mining system 200 can use the encoder neural network 220, which has been trained as part of a larger prediction neural network 240 trained to generate object detection prediction data, to generate a feature vector 222 for each of one or more regions within the LiDAR sensor input that are candidates for depicting an object”), and a linguistic feature of the input feature ([0085] The REM query system 600 can generate a textual embedding for the text 604…”)
selecting a target image feature corresponding to the linguistic feature from among the candidate image features ([0084] “One or more embeddings (referred to below as “query embeddings”) will be generated by the prediction neural networks during the processing of the sensor inputs 602 to output the rareness scores 612. The REM query system 600 thus can use a selection engine 620 to select, from the library of embeddings 630, one or more similar embeddings to the query embeddings generated from the sensor inputs 602 that are referenced in the query received by the system 600”) based on similarity scores of similarities between the candidate image features and the linguistic features ([0103] “The system selects one or more similar embeddings based on similarities of the embeddings with respect to the sensor input referenced in the query (step 904). Here, “similarity” is defined in terms of a distance in an embedding space between the query embedding and each of the plurality of embeddings maintained by the system…The distance may be computed in any appropriate way, such as with Euclidean distance, Hamming distance, cosine similarity, or the like”)
generating a [0047] configured to encode multiple modalities – point cloud, camera, etc) based on the target image feature and the point cloud feature (Fig. 2, 242 “Intermediate Feature Map”; [0051])
detecting a 3D bounding box corresponding to the target object by executing an object detection model based on the [0007] “FIG. 4 is an example illustration of a prediction output for the LiDAR sensor input of FIG. 3”; [0062] “FIG. 4 is an example illustration of a prediction output for the LiDAR sensor input of FIG. 3. The prediction output includes object detection prediction data generated by a prediction neural network. The object detection prediction data includes a 3-D bounding box encompassing each detected object in the LiDAR point cloud. The bounding box may be rectangular, oval, or any other shape. As shown in FIG. 4 , the object detection prediction data includes bounding box 401 for object 301, bounding box 402 for object 302, bounding box 403 for object 303, bounding box 404 for object 304, bounding box 405 for object 305, and bounding boxes 406 and 407 for object 306 (one corresponding to the cab of the truck, and one corresponding to the trailer of the truck)”)
Zhao does not explicitly teach:
a decoding output
However, Yang, in the same field of endeavor of point cloud object detection, teaches:
a decoding output (Fig. 3, fully decoded output of 335 feeds into “Bounding Box Refinement Model” 348; [0052] “The pose residuals (344) are collections of data generated by the decoder model (335) from the updated feature vectors (332). The pose residuals (344) are the differences between observed values and predicted values for location and direction information. For example, one of the pose residuals (344) is the difference between one of the bounding box vectors (310) and a predicted value for the bounding box vector. Such that adding a pose residual to the bounding box vector may yield the predicted value for the location and direction of the bounding box vector.”; [0054])
It would have been obvious to one of ordinary skill in the art as of the effective filing date of the claimed invention, to modify Zhao to include the limitations of decoded bounding box input, as taught by Yang. Under the broader teachings of Zhao, it is reasonable to assume the more broadly disclosed “Prediction Neural Network” 240 of Zhao’s Fig. 2, could reasonably include a decoder for the generation of “Prediction Output” 244. Yang further evidences this possibility, while also further suggesting the advantage of an output with increased refinement.
With respect to claim 2, Zhao and Yang teach the method of claim 1.
Zhao further teaches:
generating the linguistic feature corresponding to the input language by a language encoding model performing inference on the input language ([0085] “Additionally or alternatively, the REM query system 600 can receive a query that includes text 604 and use the selection engine 620 to retrieve one or more sensor inputs 624 based on the text 604. For example, the text 604 may include phrases of one or more terms that are descriptive of user-specified contents that should appear in sensor inputs. The REM query system 600 can generate a textual embedding for the text 604 and then select, for the textual embedding and from the library of embeddings 630, one or more embeddings that are similar to the textual embedding for the text 604.”)
generating candidate image features corresponding to partial areas of the input image by executing an image encoding model ([0047] “For each sensor input 211 included in the dataset 210, the system 200 uses the encoder neural network 220 to generate one or more feature vectors for the sensor input.”) and a region proposal [0050] “For example, one or more feature vectors 222 can be generated from an intermediate feature map of a sensor input 211 by cropping the intermediate feature map (e.g., using region of interest (ROI) pooling), applying a predetermined sequence of transformations, and so on”; [0089] “The system processes the sensor input using an encoder neural network to generate one or more feature vectors for the sensor input (step 704). For example, the system can generate a feature vector for a region in the sensor input, an object depicted in the region, or both")
generating a point cloud feature corresponding to the input point cloud by executing Fig. 2; [0047])
Zhao does not explicitly teach:
a region proposal model
a point cloud encoding model
As the feature extraction is broadly understood under Fig. 2 to be performed by the “Encoder Neural Network” rather than separate models per se.
However, Yang teaches:
a region proposal model (Fig. 3, “Bounding Box Refinement Model” 348)
generating a point cloud feature corresponding to the input point cloud by executing a point cloud encoding model based on the input point cloud (Fig. 3, “Course Trajectory Model” 308; [0035])
It would have been obvious to one of ordinary skill in the art as of the effective filing date of the claimed invention, to modify Zhao to include the separate modalities of Yang. Doing so would have the advantage of greater compartmentalizing and optimizing functions individually under different models. The systems readily integrate, as it would involve merely dividing Zhao’s “Encoder Neural Network” into three separate modalities.
With respect to claim 10, Zhao and Yang teach the method of claim 1.
Zhao further teaches:
A non-transitory computer-readable storage medium storing instructions that, when executed by a processor, cause the processor to perform the method of claim 1 ([0125]-[0126]).
Yang further teaches:
A non-transitory computer-readable storage medium storing instructions that, when executed by a processor, cause the processor to perform the method of claim 1 ([0057])
With respect to claim 11, it is functionally the same as claim 1, but recited as an electronic device comprising a processor, and memory storing instructions configured to cause the processor to perform the method of claim 1. These additional hardware limitations are taught by both Zhao ([0125]-[0126]) and Yang ([0057]). Accordingly, claim 11 is rejected in line with the analysis above.
With respect to claim 12, it is functionally the same as claim 2, but recited as functions of the electronic device of claim 11. Zhao and Yang teach the electronic device, as discussed above. Accordingly, claim 12 is rejected.
With respect to claim 20, it is functionally similar to claim 1, but recited as functions of a vehicle comprising:
a camera configured to generate an input image with respect to a three-dimensional (3D) space
a light detection and ranging (lidar) sensor configured to generate an input point cloud with respect to the 3D space
one or more processors configured to [perform the method of claim 1]
a control system configured to control the vehicle based on the 3D bounding box
Zhao further teaches:
a vehicle ([0003]) comprising:
a camera configured to generate an input image with respect to a three-dimensional (3D) space ([0031]; [0088])
a light detection and ranging (lidar) sensor configured to generate an input point cloud with respect to the 3D space ([0031; [0088]
one or more processors configured to [perform the method of claim 1]
Zhao does not explicitly teach:
a control system configured to control the vehicle based on the 3D bounding box
However, Yang teaches:
a control system configured to control the vehicle based on the 3D bounding box (Fig. 1; [0021] “The autonomous system (116) includes a virtual driver (102) that is the decision making portion of the autonomous system (116). The virtual driver (102) is an artificial intelligence system that learns how to interact in the real world and interacts accordingly.”; Fig. 3 (all bounding box outputs within larger “Virtual Driver 102” system))
It would have been obvious to one of ordinary skill in the art as of the effective filing date of the claimed invention, to modify Zhao to include the limitations of autonomous vehicle driving based on bounding box generation, as taught by Yang. Doing so has the advantage of integrating the system as part of an AI driver for downstream operation. The systems readily integrate, as the system of Zhao is already embodied within an autonomous vehicle ([0003]).
The remaining limitations are rejected as taught and modified in line with the analysis of claims 11 and 1.
Claims 3-5 and 13-15 are rejected under 35 U.S.C. 103 as being unpatentable over Zhao and Yang in view of Shafiullah et. al CLIP-Fields: Weakly Supervised Semantic Fields for Robotic Memory (Hereinafter, “Shafiullah”)
With respect to claim 3, Zhao and Yang teach the method of claim 1.
Zhao further teaches the limitations of:
generating extended expressions Fig. 8; [0099] “The system provides the rareness scores in response to receiving the query (step 812), e.g., to the user that submitted the query)
Zhao does not explicitly teach:
each comprising (i) a position field that indicates a geometric characteristic of the target object based on the input language and comprising (ii) a class field indicates a class of the target object, and
wherein the linguistic feature is generated based on the extended expression
However, Shafiullah, in the same field of endeavor of semantic object detection, teaches:
generating extended expressions each comprising (i) a position field that indicates a geometric characteristic of the target object (Fig. 2, noting that identified “coffee maker” has distinct outline relative to other objects. Read in line with [0054] of the claimed invention’s specification, “The geometric characteristic of the position field 2111 may include a geometric position, a geometric shape, and/or the like”) and comprising (ii) a class field indicates a class of the target object (Fig. 2, noting labels for the objects generated)
wherein the linguistic feature is generated based on the extended expression (Fig. 2, a linguistic feature is generated by “Sentence-BERT” using as input the labels generated; [4B] “In both cases we derive a set of detected objects with language labels in the image, along with their label masks and confidence scores. We back-project the pixels included in the label mask to the world coordinates using our point cloud. We label each back-projected point in the world with the associated language label and label confidence score”. In other words, a linguistic feature is generated (labels in 3D) based on the extended expression (label mask in 2D); Fig. 3)
It would have been obvious to one of ordinary skill in the art as of the effective filing date of the claimed invention, to modify Zhao and Yang to include the limitations of positions fields, class fields, and linguistic feature generation as taught by Shafiullah.
Including position fields based on text input would have the advantage of providing further information to the user about the geometry of a queried object. This also comports with the objective of Zhao, which seeks to provide a user with useful information about an object in response to a text input. Additionally, in a separate embodiment of Shafiullah, the authors allow a robot to navigate to different objects using text queries and their computer vision model ([5B(3)] “Next, on our robot, we load the CLIP-Field to help with the localization and navigation of the robot…” wherein CLIP-Field is depicted in Fig. 11 with according generated separate-embodiment “position fields” responsive to queries, “stored representations” in the form of masks). Accordingly, one of ordinary skill in the art would understand that a text query could also output such a “position field” directly.
Similarly, one of ordinary skill in the art would appreciate the advantage of class labels as additionally generated extended expressions. Class labels provide further feedback about an object’s identity, in addition to geometry. Zhao also further understands that objects will generally have labels as well ([0017] “For example, humans dressed in various types of clothing may all have the same class label of pedestrian, however certain costumes (e.g., Halloween costumes or masquerade ball costumes) may change the shape of the human’s body). Accordingly, the limitation readily integrates.
Linguistic feature generation as taught by Shafiullah involves using a first set of “extended expressions” (which include class labels) to create a second. Doing so has the advantage of translating representations across dimensional spaces. The systems readily integrate in a similar manner, as Zhao also processes both 3D and 2D sensor information.
With respect to claim 4, Zhao, Yang and Shafiullah teach the method of claim 3.
Shafiullah further teaches:
wherein objects of a same class and with different geometric characteristics are distinguished from each other based on the position fields of the extended expressions (Fig. 2 noting that objects both have different labels and also shapes (coffee maker))
With respect to claim 5, Zhao, Yang and Shafiullah teach the method of claim 3.
Shafiullah further teaches:
wherein the position field is learned through training ([4B] discussing how one can use pre-trained objection detection models like Detic to generate a position fields)
With respect to claims 13-15, they are functionally the same as claims 3-5, but recited as functions of the electronic device of claim 11. Zhao and Yang teach the electronic device, as discussed above. Accordingly, claims 13-15 are rejected.
Claims 6-8 and 16-18 are rejected under 35 U.S.C. 103 as being unpatentable over Zhao and Yang in view of Wang et. al CRIS: CLIP-Driven Referring Image Segmentation (Hereinafter, “Wang”)
With respect to claim 6, Zhao and Yang teach the method of claim 1.
Zhao and Yang further teach:
generating image tokens by segmenting the target image feature (Zhao, [0050] “For example, one or more feature vectors 222 can be generated from an intermediate feature map of a sensor input 211 by cropping the immediate feature map (e.g., using region of interest (ROI) pooling…)
generating point cloud tokens by segmenting the point cloud feature (Yang, Fig. 3, “Box Encoder Model” 318, inputting into “Combination Model” 322; Yang, [0016] “A combined feature vector may be analogous to the vector generated from a token that represents a word from a sentence and is input to a language model…”; Zhao, [0050])
generating first position information indicating relative positions of the respective image tokens (Zhao, [0052] “As another example, if the prediction neural network 240 is trained to generate as prediction output 244 agent behavior prediction, then the intermediate feature 242 may contain information that characterizes, for an agent depicted in the sensor input 211, a motion of the agent given the environmental context, e.g., given some or all of the states of the environment when the motion occurred. For example, the intermediate feature map 242 may contain information that characterizes the motion of the agent relative to another agent, a spatial relationship of environment objects to the agent, or the like”)
generating second position information indicating relative positions of the respective point cloud tokens (Zhao, [0052] noting tracking “motion of the agent relative to other agent”; Yang, [0046] “The perturbations may include changing the positions, sizes, and headings identified by the bounding box vectors within the training data”)
executing the Yang, [0063]; Yang, Fig. 3, 328; Zhao, [0050]; Zhao, [0052]; Zhao, Fig. 2, 240)
Zhao and Yang do not explicitly teach:
the multi-modal decoding model
(As in the context of the combined “multi-modal decoding model” teaching discussed in claim 1, execution of attention has already been performed. Yang executes key data and value data attention with an encoder (Fig. 3, 328; [0063]), prior to passing resultant vectors to Zhao/Yang’s “multi-modal decoder model” (Fig. 3, 335))
key data and value data based on the image tokens
However, Wang in the same field of endeavor of 3D segmentation, teaches:
generating image tokens (Fig. 3, “Visual Token”)
generating first position information indicating relative positions of the respective image tokens (Fig. 3, “Position Encoding”)
executing the multi-modal decoding model ([3.2]; Fig. 3 “Vision-Language Decoder” x n) with key data and value data based on the image tokens, [3.2])
PNG
media_image1.png
656
496
media_image1.png
Greyscale
PNG
media_image2.png
73
483
media_image2.png
Greyscale
It would have been obvious to one of ordinary skill in the art, as of the effective filing date of the claimed invention, to modify Zhao and Yang to include the limitations of image token attention and multi-modal decoding, as taught by Wang. Doing so would have the advantage of enhancing accuracy by considering global context and relationships between pixels. Integrating attention processing through the decoder rather than the encoder also has the advantage of greater reconstructing fine-grained details. The systems readily integrate, as Zhao already collects the necessary information to perform such processing through its sensor input encoding and prediction neural networks. Yang further teaches attention mechanisms, as integrated within the combined multi-modal decoding model of Zhao and Yang. One of ordinary skill in the art would understand that the general attention mechanism taught could be performed both in encoded-feature space, and/or by a later decoding block (integrated as either part of Yang’s decoder model (Fig. 3, 335) or an intervening block before).
With respect to claim 7, Zhao, Yang, and Wang teach the method of claim 6.
Zhao, Yang, and Wang further teach:
wherein the generating of the decoding output further comprises executing the multi-modal decoding model with query data (Wang, [3.2] discussing queries; Yang, [0063] “In an embodiment, executing the attention model includes executing an attention layer of the attention model to perform one or more transformations to the set of combined feature vectors…Self-attention may be used with multiple sets of query, key, and value matrices (i.e., multiple heads) for each layer to generate output vectors for the layers”; Yang, Fig. 3, 328; Zhao, [0050]; Zhao, [0052]; Zhao, Fig. 2, 240) based on detection guide information indicating detection position candidates with a possibility of detecting the target object in 3D space (Yang, understanding that the initial vectors and trajectories generated from sensor data are estimates of object position and motion that are refined throughout execution of the model in Fig. 3; Yang [0046] “The perturbations may include changing the positions, sizes, and headings identified by the bounding box vectors within the training data”; Yang Fig. 5; Yang [0045] “In an embodiment, a set of one or more of the combined feature vectors (325) correspond to a trajectory of an object that includes a set of the bounding box vectors”; Yang [0018] “The pose residuals and size residuals may be related to and quantify the error between the original location estimates from sensory data and the actual location of objects that the self-driving system identifies and avoids”; Zhao teaching “detection guide information” more broadly while not being part of query input, [0033] “For example, the prediction data 114 can define a plurality of bounding boxes with reference to the environment characterized by the sensor data 110 and, for each of the plurality of bounding boxes, a respective likelihood that an object belonging to an object category from a set of possible object categories is present in the region of the environment shown in the bounding box.”)
With respect to claim 8, Zhao, Yang, and Wang teach the method of claim 7
Yang further teaches:
wherein the detection position candidates are distributed non-uniformly (Fig. 5, with the understanding that a point cloud received from a LiDAR sensor and resultant vectors are “non-uniform” under the BRI of the term (i.e., having a degree of variation rather than conforming to a rule or standard); Yang [0022] “A real-world environment is the portion of the real world through which the autonomous system (116), when trained, is designed to move. Thus, the real-world environment may include concrete and land, construction, and other objects in a geographic region along with agents. The agents are the other agents in the real-world environment that are capable of moving through the real-world environment. Agents may have independent decision making functionality. The independent decision-making functionality of the agent may dictate how the agent moves through the environment and may be based on visual or tactile cues from the real-world environment. For example, agents may include other autonomous and non-autonomous transportation systems (e.g., other vehicles, bicyclists, robots), pedestrians, animals, etc.”; [0025] “In order to interact with the real-world environment, the autonomous system (116) includes various types of sensors...which are used to obtain measurements of the real-world environment” (as opposed to a pre-determined and understood environmental/synthetic data). The examiner understands the general disclosure of the claimed invention’s specification to include indications of non-uniformity within the detection position information, however, it would be improper to read this into the claim as a limitation without an appropriate recitation)
PNG
media_image3.png
790
1055
media_image3.png
Greyscale
With respect to claims 16-18, they are functionally the same as claim 6-8, but recited as functions of the electronic device of claim 11. Zhao and Yang teach the electronic device, as discussed above. Accordingly, claims 16-18 are rejected.
Claims 9 and 19 are rejected under 35 U.S.C. 103 as being unpatentable over Zhao, Yang and Wang in view of Frossard et. al End-to-end Learning of Multi-sensor 3D Tracking by Detection (Hereinafter, “Frossard”)
With respect to claim 9, Zhao, Yang, and Wang teach the method of claim 7.
Zhao, Yang, and Wang do not explicitly teach:
wherein the multi-modal decoding model generates the decoding output by extracting a correlation from the target image feature, the point cloud feature and the detection guide information.
(Though broadly it is understood that Zhao is configured to fuse image and LiDAR inputs [0044], and Yang teaches detection guide information as a possible feature as well)
However, Frossard, in the same field of endeavor of 3D object tracking and detection, teaches:
wherein the multi-modal decoding model generates the decoding output by extracting a correlation from the target image feature, the point cloud feature and the detection guide information (Fig. 1, image and 3D LiDAR inputs are correlated to “detection guide information” (determining trajectories of “candidate detections” with respect to other “candidate detections” across frames [3A]-[3B]; Fig. 1 “Detection Net”), then are correlated to each other, [3C(1)])
PNG
media_image4.png
439
644
media_image4.png
Greyscale
PNG
media_image5.png
605
1373
media_image5.png
Greyscale
It would have been obvious to one of ordinary skill in the art as of the effective filing date of the claimed invention, to modify Zhao, Yang, and Wang to include the limitations of multi-modality fusion with respect to detection guide information, as taught by Frossard. Doing so would have the advantage of confirming and verifying the existence of candidate detections and trajectories across multiple sensor inputs. The teachings of Frossard readily integrate with the system of Zhao, Yang, and Wang. Zhao is already configured to fuse sensor inputs in LiDAR and image spaces. Yang and Zhao are both capable of tracking candidate objects/trajectories. Wang also encourages correlation amongst modalities generally through the use of a decoder, though across image and text modalities rather than 3D and 2D.
With respect to claim 19, it is functionally the same as claim 9, but recited as functions of the electronic device of claim 11. Zhao and Yang teach the electronic device, as discussed above. Accordingly, claim 19 is rejected.
Additional References
Additionally cited references (see attached PTO-892) otherwise not relied upon above have been made of record in view of the manner in which they evidence the general state of the art.
The examiner draws attention to Kerr et. al LERF: Language Embedded Radiance Fields (Hereinafter, “Kerr”), with respect to claim 3. Kerr further evidences the generation of an extended “position field” expressions, indicating geometric characteristics of a target object in response to textual prompts (Fig. 3).
The examiner also draws attention to Liu et. al Segment Any Point Cloud Sequences by Distilling Vision Foundation Models (Hereinafter, “Liu”) with respect to the claimed invention broadly, and claim 9 specifically. Liu’s teachings include a correlation of 2D and 3D LiDAR data points across multiple frames (Fig. 2; [3.2]).
Inquiry
Any inquiry concerning this communication or earlier communications from the examiner should be directed to NOAH WILLIAM BOYAR whose telephone number is (571)272-8392. The examiner can normally be reached 8:30 – 5:00 EST, Monday – Friday.
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, Chan Park can be reached at 571-272-7409. 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.
/NOAH W BOYAR/Examiner, Art Unit 2669
/CHAN S PARK/Supervisory Patent Examiner, Art Unit 2669
Prosecution Insights