In this paper we tackle the problem of vehicle re-identification in a camera network utilizing triplet embeddings. Re-identification is the problem of matching appearances of objects across different cameras. With the proliferation of surveillance cameras enabling smart and safer cities, there is an ever-increasing need to re-identify vehicles across cameras. Typical challenges arising in smart city scenarios include variations of viewpoints, illumination and self occlusions. Most successful approaches for re-identification involve (deep) learning an embedding space such that the vehicles of same identities are projected closer to one another, compared to the vehicles representing different identities. Popular loss functions for learning an embedding (space) include contrastive or triplet loss. In this paper we provide an extensive evaluation of these losses applied to vehicle re-identification and demonstrate that using the best practices for learning embeddings outperform most of the previous approaches proposed in the vehicle re-identification literature. Compared to most existing state-of-the-art approaches, our approach is simpler and more straightforward for training utilizing only identity-level annotations, along with one of the smallest published embedding dimensions for efficient inference. Furthermore in this work we introduce a formal evaluation of a triplet sampling variant (batch sample) into the re-identification literature.

With an exponential growth in the use of digital documents and photographic and medical images, the interest in lossless image compression has increased. Coders, such as CALIC and JPEG-LS, using context modeling have raised the bar on achievable compression performance. However, the computation required for these coders is significant and naturally serial. Parallelizable and compute efficient compression and decompression algorithms have attractive features such as cost effective hardware and scalable software implementations. Hence, we propose a compute friendly, parallel algorithm for image compression with compression performance comparable to that of the state-of-the-art schemes. Furthermore, we also show that the proposed method is applicable to interframe lossless coding for image sequences such as medical, graphical, and video contents.

The newly introduced Microprocessor Architecture for Java Computing (MAJC) supports parallelism in a hierarchy of levels: multiprocessors on chip, vertical micro threading, instruction level parallelism via a very long instruction word architecture (VLIW) and SIMD. The first implementation, MAJC 5200, includes some key features of MAJC to realize a high performance multimedia processor. Two CPUs running at 500 MHz are integrated into the chip to provide 6.16 GFLOPS and 12.33 GOPS with high speed interfaces providing a peak input-output (I/O) data rate of more than 4.8 GBytes/second. The chip is suitable for a number of applications including graphics/multimedia processing for high-end set-top boxes, digital voice processing for telecommunications, and advanced imaging.

We propose a new technique of rate control for very low bit rate video coding for visual telephony. The scheme is optimized for a video coder conforming to the H.263 standard and outperforms the linear rate control scheme used in the test model in terms of visual performance and frame jerkiness at the output. The rate control scheme,is designed for a variable bit-rate and variable frame rate coder. We address the rate allocation problem as a constrained optimization problem and exploit the syntax of H.263 to use a constrained trellis to perform the optimization. Furthermore, we reduce the computations in the trellis search by modeling the rate distortion functions. The resulting bit allocation yields a higher signal to noise ratio, and a better correlation between the quantization step size and the block energies, with respect to the linear rate control scheme for the same number of bits assigned for a particular frame.

The discrete wavelet transform has recently emerged as a powerful technique for decomposing images into various multi-resolution approximations. Multi-resolution decomposition schemes have proven to be very effective for high-quality, low bit-rate image coding. In this work, we investigate the use of entropy-constrained trellis-coded quantization (ECTCQ) for encoding the wavelet coefficients of both monochrome and color images. ECTCQ is known as an effective scheme for quantizing memoryless sources with low to moderate complexity, The ECTCQ approach to data compression has led to some of the most effective source codes found to date for memoryless sources. Performance comparisons are made using the classical quadrature mirror filter bank of Johnston and nine-tap spline filters that were built from biorthogonal wavelet bases. We conclude that the encoded images obtained from the system employing nine-tap spline filters are marginally superior although at the expense of additional computational burden. Excellent peak-signal-to-noise ratios are obtained for encoding monochrome and color versions of the 512x512 "Lenna" image. Comparisons with other results from the literature reveal that the proposed wavelet coder is quite competitive.

Investigates the performance of different adaptive prediction algorithms for use in a low delay, high-quality speech coder at 18 kbit/s. The system to be studied is based on trellis coded quantization (TCQ). In the work presented, the authors demonstrate that when trellises based on feedback-free convolutional encoding circuits are employed, error propagation in TCQ systems is not a serious problem. Performance comparisons are made using a fixed predictor and four different adaptive predictors. Informal listening tests reveal that the encoded speech is of good communications quality at an error rate of 10/sup -3/. In an ideal channel, a system employing an all-pole least squares lattice predictor with pitch prediction produces reconstructed speech that is nearly indistinguishable from the original.>

The application of transform coding using trellis coded quantization (TCQ) to encode monochrome and color images is investigated. Specifically, TCQ is used to encode transform coefficients resulting from applying a 16 * 16 discrete cosine transform (DCT) to 8-b gray level and 24-b color images. For the color images, the red, green, and blue (RGB) planes are transformed into the NTSC transmission primaries (Y,I,and Q) before the DCT is applied. Rate allocation schemes for encoding the transform coefficients and side information are presented and discussed. The generalized Lloyd algorithm and the entropy-constrained vector quantizer design algorithm are used to design fixed-rate and entropy-constrained codebooks. The entropy-constrained designs result in a low-complexity system with objective performance rivaling that of many coding schemes from the literature.>