A participant (or a team) can submit to and win prizes in either or both categories. The submissions will be a single model in TensorflowLite format (https://www.tensorflow.org/lite/).
Our evaluation server during the competition will use a fixed Github commit (4b2cb67756009dda843c6b56a8b320c8a54373e0)
Please only use the servers as a tool for final verification. Consider using the resources (validator binary, local benchmarker app, latency tables, sample models) in Section 4 for fast iterations on your local development.
Participants are encouraged to check out public implementations of architecture search methods such as ProxylessNAS.
_____________________________________________________________________________________________________________________________
1. Object Detection Challenge (CPU)
_____________________________________________________________________________________________________________________________
Both training and evaluation data are from the COCO object detection challenge. Competition will provide an online leaderboard showing model performance on the minival dataset (Image Ids found here). Final evaluation will use 2017 test-dev. Note that discrepancies are expected between these two metrics.
Submissions are evaluated based on task-performance (which is mAP for object detection and top-1 accuracy for image classification) under a targeted average latency.
For object detection, the metric is COCO mAP and the targeted latency is 30ms per image.
The performance improvement will be measured relative to the empirical Pareto frontier, established by previous OVIC submissions as well as the Mobilenet family. Specifically, we estimate the Pareto frontier curve in the following parametric form:
a(t) = k log(t) + a0
where t is the average latency of the model in milliseconds, a(t) is the best task-performance at latency t in percent, k and a0 are parameters.
We retrain the Pareto frontier periodically to adjust for codebase changes. The frontier is different each time but the differences are small.
The latest (Apr 7, 2020) estimate of the Pareto frontier for the miniVal set is: k =16.894553358968146, a0 = -34.42191514521174.
This Pareto frontier, estimated from all submissions prior to CVPR2020, shall be used to assess and compare the quality of the submissions at different latencies. Note that this may be high due to participants unintentionally training on the miniVal set.
Given a submission with performance A and latency T, the metric M of the submission is:
If T is within +/- 20% of the latency target:
M(A, T) = A - a(T)
If T is faster than 80% of the latency target, it’ll be treated as if the latency is 80% of the latency target.
If T is slower than 120% of the latency target, the submission is deemed invalid.
Figure above illustrates Pareto frontiers estimated from baseline models (shown s dots) and how the evaluation metric is computed for a submission (star) as the offset from the estimated frontier. Illustration only, no real data point used.
The submissions will be interpreted using the latest version of TensorFlow Lite and benchmarked using a single thread with a batch-size of 1 on a single big core of the Pixel 4 smartphone.
As the TensorFlow Lite version may change during the competition period, the evaluation server will update frequently and re-measure the latency for all submissions. The best score (across all server builds) for each submission will be used towards the final scoring for that submission.
The models must expect input tensors with dimensions [1 x input_height x input_width x 3], where the first dimension is batch size and the last dimension is channel count, and input_height and input_width are the integer height and width expected by the model, each must be between 1 and 1000.
All images will be resized to these dimensions, and they should be picked judiciously to balance task-performance and latency.
Inputs should contain RGB values between 0 and 255. To see how the images are processed before feeding into the TensorFlow Lite model, check out the convertBitmapToInput function from the SDK.
The output should contain four tensors:
The recommended way to produce these tensors is to use Tensorflow’s object detection API. Let config_path points to the TrainEvalPipelineConfig used to create and train the model, and checkpoint_path points to the checkpoint of the model.
Participants can create a frozen tensorflow model in directory output_dir using the following command:
bazel-bin/tensorflow_models/object_detection/export_tflite_ssd_graph \ --pipeline_config_path="${config_path}" --output_directory="${output_dir}" \ --trained_checkpoint_prefix="${checkpoint_path}" \ --max_detections=100 \ --add_postprocessing_op=true \ --use_regular_nms=${use_regular_nms} |
Where use_regular_nms is a binary flag that controls whether the regular non-max suppression is used, with the alternative being a faster non-max suppression implementation that is less accurate.
Participants can convert their Tensorflow model into a submission-ready model using the following command:
bazel-bin/tensorflow/lite/python/tflite_convert \ --graph_def_file="${local_frozen}" --output_file="${tflite_file}" \ --input_format=TENSORFLOW_GRAPHDEF --output_format=TFLITE \ --inference_type=${inference_type} \ --input_shapes="1,${input_height},${input_width},3" \ --input_arrays="${input_array}" \ --output_arrays=\ 'TFLite_Detection_PostProcess',\ 'TFLite_Detection_PostProcess:1',\ 'TFLite_Detection_PostProcess:2',\ 'TFLite_Detection_PostProcess:3' \ --nochange_concat_input_ranges --noallow_custom_ops \ --mean_values="${mean_value}" --std_dev_values="${std_value}" |
where local_frozen is the frozen graph definition;
inference_type is either FLOAT or QUANTIZED_UINT8;
input_array and output_array are the names of the input and output in the tensorflow graph; and mean_value and std_dev_value are the mean and standard deviation of the input image.
Note that:
The input type is always QUANTIZED_UINT8, and specifically, RGB images with pixel values between 0 and 255. This requirement implies that for floating point models, a Dequantize op will be automatically inserted at the beginning of the graph to convert UINT8 inputs to floating-point by subtracting mean_value and dividing by std_dev_value.
_____________________________________________________________________________________________________________________________
2. Image Classification Challenge (CPU)
_____________________________________________________________________________________________________________________________
Training data are from ImageNet classification dataset available at the ILSVRC 2012 website. The evaluation data are the holdout images from the 2018 competition.
The same relative metric in Section 1.2 is used for classification. The only differences are that the task-performance is top-1 classification accuracy, and the targeted latency is 10ms per image.
The latest estimate (Apr 7, 2020) of the Pareto frontier of the validation set is: k = 49.84607103726407, a0 = -21.759878323711725.
These parameters are estimated from the CVPR2020 submissions.
See Section 1.2 and 1.3 in object detection. To see how the images are processed before feeding into the TensorFlow Lite model, check out the convertBitmapToInput function from the SDK (note that ImageNet preprocessing is different from that of COCO).
The output must be a [1 x 1001] tensor encoding probabilities of the classes, with the first value corresponding to the “background” class. The list of the full labels is here.
The participants can convert their Tensorflow model into a submission-ready model using the following command:
bazel-bin/tensorflow/lite/python/tflite_convert-- \ --graph_def_file="${local_frozen}" --output_file="${tflite_file}" --input_format=TENSORFLOW_GRAPHDEF --output_format=TFLITE \ --inference_type="${inference_type}" \ --inference_input_type=QUANTIZED_UINT8 \ --input_shape="1,${input_height},${input_width},3" \ --input_array="${input_array}" \ --output_array="${output_array}" \ --mean_value="${mean_value}" --std_dev_value="${std_dev_value}" |
See Section 1.5 of object detection for the definition of the arguments. _____________________________________________________________________________________________________________________________
Step 3 Image Classification Challenge (DSP) _____________________________________________________________________________________________________________________________
Training data are from ImageNet classification dataset available at the ILSVRC 2012 website. The evaluation data are holdout images collected following the ImageNet protocol.
_____________________________________________________________________________________________________________________________
_____________________________________________________________________________________________________________________________
The primary resource is the OVIC benchmarker page (https://github.com/tensorflow/tensorflow/tree/master/tensorflow/lite/java/ovic). It contains:
Participants are encouraged to check out the tutorials for post-training quantization (https://www.tensorflow.org/lite/performance/post_training_quantization) and quantization-aware training (https://www.tensorflow.org/model_optimization/guide/quantization/training) for training quantized Mobilenet models.
Participants are also encouraged to check out Tensorflow’s ObjectDetectionAPI tutorial (https://github.com/tensorflow/models/blob/master/research/object_detection/README.md) for training detection models. _____________________________________________________________________________________________________________________________
Disclaimers:
_____________________________________________________________________________________________________________________________
All participants must be at least 13 years old, not a citizen of US embargoed countries, and not affiliated with the organizers or sponsors (Purdue University, Duke University, University of North Carolina Chapel Hill or employees Alphabet Inc).
All submissions, along with the empirical Pareto frontier, will be re-computed after submission closes using the same codebase version. Regressions / improvements may happen as a result of versioning differences between the time of submission and the time of evaluation. In case of a significant regression the organizers may consider using the better measurement between the two.